Conference PaperPDF Available

Risk Management with Hard-Soft Data Fusion in Maritime Domain Awareness


Abstract and Figures

Enhanced situational awareness is integral to risk management and response evaluation. Dynamic systems that incorporate both hard and soft data sources allow for comprehensive situational frameworks which can supplement physical models with conceptual notions of risk. The processing of widely available semi-structured textual data sources can produce soft information that is readily consumable by such a framework. In this paper, we augment the situational awareness capabilities of a recently proposed risk management framework (RMF) with the incorporation of soft data. We illustrate the beneficial role of the hard-soft data fusion in the characterization and evaluation of potential vessels in distress within Maritime Domain Awareness (MDA) scenarios. Risk features pertaining to maritime vessels are defined a priori and then quantified in real time using both hard (e.g., Automatic Identification System, Douglas Sea Scale) as well as soft (e.g., historical records of worldwide maritime incidents) data sources. A risk-aware metric to quantify the effectiveness of the hard-soft fusion process is also proposed. Though illustrated with MDA scenarios, the proposed hard-soft fusion methodology within the RMF can be readily applied to other domains.
Content may be subject to copyright.
Risk Management with Hard-Soft Data Fusion
in Maritime Domain Awareness
Rafael Falcon, Rami Abielmona, Sean Billings, Alex Plachkov and Hussein Abbass
Abstract—Enhanced situational awareness is integral to risk
management and response evaluation. Dynamic systems that
incorporate both hard and soft data sources allow for compre-
hensive situational frameworks which can supplement physical
models with conceptual notions of risk. The processing of widely
available semi-structured textual data sources can produce soft
information that is readily consumable by such a framework. In
this paper, we augment the situational awareness capabilities of a
recently proposed risk management framework (RMF) with the
incorporation of soft data. We illustrate the beneficial role of the
hard-soft data fusion in the characterization and evaluation of
potential vessels in distress within Maritime Domain Awareness
(MDA) scenarios. Risk features pertaining to maritime vessels are
defined a priori and then quantified in real time using both hard
(e.g., Automatic Identification System, Douglas Sea Scale) as well
as soft (e.g., historical records of worldwide maritime incidents)
data sources. A risk-aware metric to quantify the effectiveness of
the hard-soft fusion process is also proposed. Though illustrated
with MDA scenarios, the proposed hard-soft fusion methodology
within the RMF can be readily applied to other domains.
Maritime Domain Awareness (MDA) can be understood
as the situational knowledge of physical and environmental
conditions that exist within or influence a maritime region. The
intended scope of this awareness includes all behaviors that
could, directly or indirectly, affect the security of the region,
its economic activity or the local environment [1].
Accurate situational awareness requires the integration of
multiple types of information that are extracted from different
data sources. High-Level Information Fusion (HLIF), defined
as Level 2 Fusion and above in the Joint Director of Labo-
ratories (JDL)/Data Fusion Information Group (DFIG) model
[2] [3], has proven to be a useful tool for the characterization
of vital MDA processes and behaviours such as anomaly
detection,trajectory prediction,intent assessment, and threat
assessment. HLIF techniques are much better positioned to
cope with the overwhelming volume of the data from a variety
of sources that flows to the maritime operations center at in-
creasingly high velocity with sometimes questionable veracity.
Maritime operators often rely on hard data sources (i.e.,
structured, quantitative, more objective, usually sensed data)
R. Falcon and R. Abielmona are with the Research & Engineering
Division, Larus Technologies Corporation, Ottawa ON, K1P 5V5 Canada
S. Billings and A. Plachkov are with the School of Electrical Engineering
& Computer Science, University of Ottawa, Ottawa ON, K1N 6N5 Canada
{sbill042, aplac099}
H. A. Abbass is with the University of New South Wales at
the Australian Defence Force Academy, Canberra, Australia ACT 2600
generated by vessel traffic in order to identify suspicious
events at sea. However, a wealth of relevant information can
be extracted from soft data sources (i.e., unstructured/semi-
structured, more subjective, qualitative data, such as textual
reports on vessel sightings or marine incidents). As demon-
strated in [4], Natural Language Processing (NLP) methods
can draw meaningful information that is representative of
human intuition, which is often not captured by hard data
sources. These pieces of soft information can then supplement
the existing hard information in order to provide a more
comprehensive situational awareness.
A risk-aware view of the system, its units and the surround-
ing environment will help identify the existing vulnerabilities
in a proactive rather than reactive fashion, which will con-
tribute to the design of effective countermeasures to mitigate
the perceived threats and assess their impact on the system.
With this goal in mind, the authors in [5] put forth a generic,
multimodular risk management framework (RMF) architecture
for real-time processing of distributed systems. The RMF
is capable of: (1) extracting a parallel risk stream from the
regular data flow periodically reported by the system units;
(2) dynamically visualizing the risk landscape of each system
unit; (3) assessing the local and global risk levels for all units
and the system as a whole and (4) producing a limited set of
promising candidate responses that can be effectuated on the
In this paper, we augment the HLIF capabilities for situa-
tional awareness (Level 2 JDL/DFIG) of the RMF in [5] with
the incorporation of soft data. We illustrate the beneficial role
of the hard-soft data fusion (HSDF) in the characterization
and evaluation of potential vessels in distress within MDA
scenarios [6]. Risk features pertaining to maritime vessels
are defined a priori and then quantified in real time using
both hard (e.g., Automatic Identification System, Douglas Sea
Scale) as well as soft (e.g., historical reports of worldwide
maritime incidents) data sources. A geographic and descriptive
representation of the incidents contained within these reports is
used to characterize the regional hostility and risk of attack for
vessels actively monitored by the RMF, hence enhancing the
overall maritime picture presented to the operator. To the best
of our knowledge, this is the first time that an RMF featuring
an HSDF component is presented and then applied to the
maritime world. Additionally, the effectiveness of the fusion
process is quantified via a newly proposed metric. Finally, the
proposed methodology is easily applicable to other domains
provided the necessary customizations are in place.
978-1-4799-5431-5/14/$31.00 c
2014 IEEE
A. The Role of Computational Intelligence (CI)
CI techniques permeate the entire RMF design (fuzzy logic
and granular computing are used for risk feature extraction;
clustering techniques for risk visualization; fuzzy inference
systems for risk assessment and evolutionary optimization
for response selection). The new fusion effectiveness metric
proposed in Section V also employs fuzzy reasoning.
B. Paper Outline
The rest of the paper is structured as follows. Section
II briefly reviews relevant works. Section III unveils the
proposed HSDF scheme for risk management and Section IV
illustrates its application within MDA. A risk-aware Measure
of Effectiveness (MoE) is put forth in Section V. Section VI is
the empirical evaluation and Section VII concludes the paper.
This Section briefly reviews several relevant studies con-
cerning maritime risk analysis, HLIF for MDA as well as
existing HSDF maritime systems.
A. Maritime Risk Analysis
The ISO 3100 [7] standard for risk management defines
risk as the effect of uncertainty on objectives. Two primary
objectives within the maritime domain are safety and secu-
rity, while uncertainty classically reflects a lack of the right
information to support the commanders intent. Consequently,
an important dimension of risk management in the maritime
domain is to provide the set of processes and tools that
support and enhance the commanders situational awareness
picture (SAP). Improving the SAP can be categorized into a
number of sub-problems, from improving the collection and
transmission of information through the fusion of data, to the
provision of intent, projection and consequence cues to support
the commanders intent. Many tools have been proposed in
the literature to provide a system-level risk picture. Hidden
Markov Models [8][9] (HMMs) are common techniques used
in this domain that rely on latent states to approximate the
dynamics of the system-level risk picture. Technically, HMMs
provide an effective method for tactical risk but can be unstable
in a dynamic environment when evaluating system-level risk.
The inter-dependency from one model to another, arising
from the interdependency of assets, makes the system stability
vulnerable when confronting a change in the environment.
Practically, the cost to sustain the integrity of large systems
composed of many interconnected HMMs is large. The same
issue arises when relying on other probabilistic models such
as Bayesian networks [10].
Complex systems research generated another line of risk
assessment methodologies for the military domain including
computational red teaming [11][12] and adversarial modelling
[13]. Practical risk-aware decision support systems have been
developed by industry such as the ATHENA Integrated De-
fense System by Raytheon [14], which is designed to search
for suspicious behaviors in the search-and-rescue division.
Two US-based projects are also worth mentioning here.
The first is the one entitled Maritime Automated Super Track
Enhanced Reporting (MASTER), which is an integrative re-
porting project supported by the Joint Capability Technology
Demonstration (JCTD) [15] and the Comprehensive Maritime
Awareness (CMA) [16] programs. The second project is the
Predictive Analysis for Naval Deployment Activities (PANDA)
[17], which is a case-based reasoning system that models
context using ontologies and business rules.
B. HLIF for Maritime Domain Awareness
Recent efforts in performing HLIF for MDA have made use
of clustering techniques to process the data sources. Clustering
is a pattern classification technique used to group similar
points; this method is widely employed due to its unsupervised
learning nature. In 2007, Laxhammar [18] tackled anomalous
vessel detection via two clustering techniques operating on Au-
tomatic Identification Systems (AIS) data, namely the Mixture
of Gaussians (MoG) model and the Fuzzy Adaptive Resonance
Theory (FuzzyART) self-organizing neural network. The data
generated by a vessel was transformed into a data point
belonging to two feature spaces: F1= (Velx, Vely) and F2
= (Velx, Vely, Lat, Lon). Both MoG and FuzzyART were
trained only on data containing normal vessel activities, and
later tested on anomalous data. The MoG is more tolerant to
noisy (anomalous) training data than FuzzyART. Due to their
unsupervised nature (which implies no need for any domain
knowledge), the two approaches are flexible enough that they
could be employed in different domains dealing with two-
dimensional motion. The limitation in the proposed solutions
is that they can detect only elementary types of anomalies (e.g.,
sea lane crossed by a vessel). The author discovered that the
inclusion of the coordinates in the second feature space did
not improve the anomaly detection capabilities.
More recently, Blasch et al. [19] explored the potential of
fusing information scattered over multiple data sets containing
hard (e.g., imagery) and soft (e.g., textual labels) data. The
authors note that the challenges at hand are mainly due to no
standardized ontological representation (i.e. no shared domain
vocabulary) used by the different data collection agencies to
describe the real-world objects contained within their data sets.
They claim that ontological alignment allows for the successful
aggregation of the information contained in the dispersed data
sets and proceed by providing a maritime example in which
the overall situational awareness was augmented by combining
vessel traffic pattern information into one fused data set.
Finally, Shao et al. [20] successfully correlated vessel con-
tacts from AIS, Synthetic Aperture Radar (SAR), and Ground
Moving Target Indicator (GMTI) data via Fuzzy K-Nearest
Neighbor classification; furthermore, the authors were able to
associate the information from the three different sources by
performing Fuzzy C-Means clustering. Lastly, they improved
vessel trajectory prediction by enhancing the performance of
the Kalman Filter in nonlinear cases with the help of an Echo
State neural network.
This Section sheds light on the extension of the RMF in
[5] and [6] with HSDF for JDL/DFIG Level 2 (situational
assessment) and Level 3 (impact assessment).
Fig. 1 displays an augmented version of the risk feature
extraction module that accounts for both hard and soft data
sources. Sensor modalities like AIS, radars, video cameras,
and weather gaugers, as well as other unsensed yet quantitative
and structured information such as oil price indicators or
dynamic country demographics are all examples of hard data
sources DS1, . . . , DSkthat can be ingested by the RMF.
Tweets, blogs, webpages, or other forms of textual reports
illustrate soft data sources DSk+1 , . . . , DSn.
To be assimilated by the RMF, the original content of these
soft data sources will have to be converted to a Common
Textual Risk Representation (CTRR). This will involve typical
NLP tasks such as tokenization, named entity recognition,
part-of-speech tagging, model representation and term an-
notation that may span both lexical and syntactic analyses.
The most important step in the CTRR transformation is the
elicitation of a risk lexicon (possibly encompassing multiple
textual sources), i.e. a dictionary of keywords that symbolizes
risk factors in the domain of interest. The soft risk feature
extractors will then use the lexicon generated by the CTRR
module to identify risk spans in the soft data, instantiate
them and use them as building blocks for the creation of soft
risk features, whose outputs must be quantitative in order to
seamlessly blend them with those of the hard risk features. The
lexicon-building process is often guided by domain experts.
Notice that in Fig. 1 the CTRR module feeds two soft risk
feature extractors: one in the object (system unit) space and
one in the response space. That is, the textual risk mined
from the soft data sources could be used to enhance the
initial characterization, provided by the hard data sources, of
an object (Level 2 JDL/DFIG, situational assessment) or a
candidate response (Level 3 JDL/DFIG, impact assessment)
needed to mitigate a perceived threat. In both spaces, the hard
risk feature extractors could choose to ingest the information
granules produced by their soft counterparts.
The rest of the RMF modules retain their original function-
ality as described in [5] and [6]. In the next section, we will
illustrate how these HSDF concepts (in the object space) are
applied to MDA scenarios.
This Section illustrates the application of the risk-aware
HSDF framework to MDA.
A. Data Sources
The following data sources were used in this study:
AIS [Hard] - A 5-day ExactEarth1feed of 227,299
AIS contacts along the northeastern coast of Canada and
USA. AIS reports include crucial information pertaining
to the vessel like its unique identifier (MMSI), type,
current location, intended course, speed, etc. We use the
vessel’s identifier, type and location in our risk-aware
fusion framework2.
SAR [Hard] - This feed contains 188 contacts from
Canada’s RADARSAT-23satellite for the same area
(northeast Canada/USA) and period of interest (POI). It
is an active tracking modality that can be used to sup-
plement AIS feeds, but it can also be used independently
to locate vessels. Given the scarce number of contacts
available, a SAR-contact-to-AIS-track association is con-
ducted as shown in [20].
Sea State Reports [Hard] - Describes the motion of
the sea waves produced by the joint effect of wind and
swell. This information is used in conjunction with the
Douglas Sea Scale4in order to compute the Sea State risk
feature. The sea data for the East Coast of Canada was
obtained from Fisheries and Oceans Canada5. To the best
of our knowledge, there is no available sea state dataset
for the Horn of Africa region, so the sea scale values
were randomly generated for that scenario.
NGA WWTTS Maritime Incident Reports [Soft] - The
Worldwide Threats to Shipping Reports (WWTTS)6is a
publicly available, semi-structured textual source main-
tained by the National Geospatial-Intelligence Agency
(NGA). Relevant maritime crime and piracy incidents
around the globe are compiled weekly and freely dissem-
inated. The textual description of each incident includes,
among other things: the kind of threat, the vessel in
question and the location where it occurred. Structured
data is extracted from this database through NLP and then
used to compute the regional hostility metric and augment
the degree of distress of a vessel [6]. Our POI goes
from 2011/01 to 2013/01, totalling 107 reports describing
2,200 incidents (732 were duplicates and hence removed).
GeoNames geographical database [Soft] - This source7
is used to extract textual repositories of city/location
names and their respective geographical coordinates to
complement NGA WWTTS incident reports where a
specific latitude and longitude is not provided.
B. MDA Scenario
An MDA scenario Sconsists of a predefined maritime area
of interest (AOI) and a group of vessels transiting through
those waters which are tracked via either passive (like AIS) or
active (e.g., SAR) sensing modalities. As a result, periodical
data about each vessel XS(e.g., position, type, heading
or speed) arrives at the maritime control center, where the
maritime operators continually monitor their statuses. This is
2AIS contacts around the Horn of Africa were synthetically generated as
we did not have access to the AIS feed of that region
4 Sea Scale
5 gdsi/waves-vagues/
6 nfpb=true& pageLabel=msi
portal page 64
Fig. 1. The RMF’s architectural blueprint showcasing hard/soft risk extractors in both the object and response spaces. Gray boxes indicate external RMF
elements. Green boxes indicate Level 2 RMF capabilities and yellow boxes indicate Level 3 RMF capabilities.
Incident Category Incident Keywords Severity Value
Bomb Threat bombed 1.0
Terrorism terrorist, terrorism 0.9
Hostage Scenario hijacked, abducted, kidnapped, hostage, kidnapping 0.8
Crew Damage fired, tied up with rope 0.7
Theft robbed, attacked, robbers, criminals, robbery, theft, stole Equipment 0.6
Invasion Boarded, clashed with, boarded the vessel, knives, invaded, trespasser 0.5
Near Invasion attempted boarding, forced, crime, threat, surrender 0.4
Threatened chased, threatened, threat, suspect, escape, blocking, risk 0.3
Approach suspicious approach, suspiciously approached, approached 0.2
Crew Error crashed, negligence 0.1
Unknown other risks 0.05
the external environment module in Fig. 1. As mentioned in
Section I, a risk stream is extracted from this original data
flow through the RMF’s risk feature extraction module.
Like in [6], maritime vessels are characterized in
terms of four risk features: (i) collision factor indicates the
likelihood of the vessel colliding with another object in the
sea; (ii) degree of distress subsumes different distress factors
such as the danger to human lives aboard the vessel, the
environmental impact of a catastrophe caused by the vessel
and the risk of running out of fuel and hence being stranded;
(iii) sea state models the distress posed by the prevalent sea
conditions (calm, moderate, rough, etc.) and is drawn from
the Douglas Sea Scale based on the vessel’s reported location
and (iv) regional hostility captures the danger of the region
where the vessel is transiting.
C. Hard-Soft Risk Feature Extraction
In the MDA scenario, a vessel Xperiodically reports its
position (latitude, longitude) and its vessel type as part of the
AIS message. The position field is used to calculate the n
closest maritime incidents to X’s current location while the
vessel type is employed to determine the similarity of X
to the other vessel types reported in the maritime incidents
repository. These two fields contribute to the calculation of
the Regional Hostility Metric for a vessel. Furthermore, the
vessel type is used to define the Risk of Attack of any vessel,
which augments the Degree of Distress risk feature.
1) Regional Hostility Metric: This risk feature, denoted by
µ(X), quantifies the perils associated with the geographical
region through which vessel Xnavigates based on historical
data in the form of maritime incident reports. Equation (1)
formalizes its modeling.
µ(X) = α M I P (X) + β MI S(X) + γ V P I (X)(1)
where αweighs the significance of X’s proximity to known
maritime incidents, βweighs the significance of the severity of
the incidents reported around the vicinity of Xand γdenotes
the significance of the pertinence of those reports to X. A
suggested weighting is α= 0.4,β= 0.3and γ= 0.3;
however, these weights could vary according to the maritime
operator’s preferences.
The mean incident proximity M IP (X)quantifies the risk
induced by the proximity of the vessel Xto reported maritime
incidents. It is defined as an average over the risk values asso-
ciated with the proximity of the nclosest maritime incidents:
M IP (X) =
where xi=dist(X.location, vi.location)is the geospatial
distance (in nautical miles, nm) from the vessel Xto the
i-th closest maritime incident and θ(xi)is termed the incident
proximity risk and defined as a fuzzy set over the domain
of distance values. Depending on the vessel’s speed, the
trapezoidal membership function that models this fuzzy set
will adopt the following parametric configurations:
A= 0; B= 0; C= 5; D= 10 for slow vessels;
A= 0; B= 0; C= 21.6; D= 43.2for medium-speed
vessels; and
A= 0; B= 0; C= 40; D= 80 for fast vessels.
The mean incident severity MI S(X)quantifies the risk
induced by the severity of the reported maritime incidents.
It is defined as an average over the risk values associated with
the severity of the nclosest maritime incidents
MIS(X) =
where yi=vi.incidentType is the type of incident in the i-th
closest report and: ψ(yi)is a mapping, termed the incident
severity risk, from an incident type to a numerical value
between 0 and 1, as shown in Table I.
The victim pertinence index V P I (X)quantifies how perti-
nent the nclosest maritime incidents are for this vessel. It is
defined as the maximum similarity between the vessel’s type
and those involved in the nclosest maritime incidents.
V PI (X) = max
i{δ(X.vesselType, vi.vesselType)}(4)
where δ(X.vesselType, vi.vesselType)is computed as fol-
δ(X.vesselT y pe, vi.vesselT y pe) =
1if X.vesselType is the same as vi.vesselType
0.5if X.vesselType is similar to vi.vesselType
0if X.vesselType is unrelated to vi.vesselType
Table II defines the similarity matrix for vessel categories
while Table III maps a vessel type to its category.
Cargo Transport Tanker/Industrial Warship Small Military Vessel Small Transport/Utility
Cargo Transport 1 0.5 0 0 0.5
Tanker/Industrial 0.5 1 0 0 0
Warship 0 0 1 0.5 0
Small Military Vessel 0 0 0.5 1 0.5
Small Transport/Utility 0.5 0 0 0.5 1
Cargo Transport Tanker/Industrial Warship Small Military Vessel Small Transport/Utility
bulk carrier chemical tanker warship coast guard boat fishing trawler
car carrier heavy lift vessel naval patrol vessel japanese harpoonists
cargo ship lpg tanker militant anti-whaling group
carrier ship mother vessel skiff
container ship oil tanker speed boat
dry cargo vessel product tanker tug boat
general cargo ship tanker dredger
livestock carrier fishing vessel
lng carrier vessel
refrigerated cargo ship
merchant ship
2) Degree of Distress: As mentioned in Section IV-B, the
Degree of Distress risk feature quantifies the combined effect
of multiple distress factors acting upon the vessel, such as
those affecting people aboard, the environment or the vessel
itself in case it runs out of fuel. In this paper, we are expanding
the formulation in [6] with another distress factor, Risk of
Attack, that will be derived from textual information mined
directly from the semi-structured NGA WWTTS maritime
incident reports, as displayed in Equation (5):
µDD (X) = 0.3µRP (X)+0.2µRE (X)+0.2µRF (X) +
where µRP (X),µRE (X)and µRF (X)are defined as in [6]
and µRA(X)is the probability that vessel Xwill be attacked
due to the category it belongs to. This value is based on
the maritime incident reports compiled in the NGA WWTTS
repository. More formally, µRA(X)is the square root of the
conditional probability that X.category would be the subject
of a maritime incident I.
µRA(X) = pP(X.category|I)(6)
where P(X.category|I)is the fraction of the total number of
reports (over a user-specified time period) where the vessel’s
category appears involved in the incident categories 1–8 listed
in Table I. Taking the square root increases the probability so
as to more uniformly distribute the Risk of Attack over [0; 1].
Despite the sound interest and rapid research developments
in the HLIF arena, the literature on MoE for HLIF is still
in its infancy. Blasch et al [21][22] were among the first
ones to propose an indicator that quantifies the degree of
effectiveness in any HLIF process. Their metric is based on
three components: (1) the information gain, measuring the
value added by the content contributed by new sources; (2)
the information quality, which reflects several facets of the
data at hand, such as its reliability and credibility and (3)
the robustness, gauging the consistency over the testing and
application domains. The authors claim that the product of
these three components is a domain-agnostic, valid strategy
for estimating HLIF effectiveness.
We formulate the information fusion effectiveness (IFE),
depicted in Equation (7), as a more tailored version of the
above general MOE by injecting risk into its components.
| {z }
Information Gain
| {z }
Information Quality
In our MDA scenario S, the Information Gain (IG) is
a measure of the number of vessels that make use of the
available data sources (Information Benefit Ratio,IBR) and
the overall risk alertness they provide (Risk Awareness Level,
RAL). More formally:
|S||{XS:X.AIS 6=∅}| for hard sources
only and
|S||{XS:Nn(X.Loc)6=∅}| for soft
sources only and
IBRSif X.AI S =∅ ∀XS
IBRHif Nn(X.Loc) = ∅ ∀XS
where |S|is the number of vessels in S,X.AIS denotes an
active AIS transceiver in vessel X,Nn(X.Loc)is the set of
the nclosest maritime incident reports to X’s current location
(see Section IV) and SRH,SRSare the relevance values of
the hard and soft data sources, respectively. In this study, we
used SRH= 0.6and SRS= 0.4
The RAL is defined as the extent to which the vessels in
the scenario Sare aware of the existing risks. This variable
is computed as the average overall risk (i.e., across all risk
features) of the vessels in Sat any time instant.
RAL =1
The Information Quality (IQ) is defined as the product of
the Information Confidence (IC) and the Overall Information
Timeliness (OIT ). I C is in turn calculated as the mean of the
product of the Reliability of Information (ROI) and Reliability
of Source (ROS) for all participating data sources:
The reliability values were defined according with the US
Military Standard 640 (MIL-STD-640). Both ROSHand
ROSShave been set to 1.0 given that the two sources
(ExactEarth for AIS and NGA for WWTTS reports) are
quite reputable private and governmental entities, respectively.
ROIHwas fixed to 0.75 (fairly reliable) since AIS can be
intentionally spoofed whereas ROISwas set to 0.85 (very re-
liable) because the maritime incident reports are peer reviewed
by knowledgeable operators prior to their online disclosure.
The OIT is a function of the Information Timeliness (IT ).
ITHcan be expressed as the average, across all vessels, of the
average delay (in seconds) between consecutive AIS reports
for each vessel. On the other hand, ITSis the average, across
all vessels, of the average age (in days) of each vessel’s n
closest maritime incident reports. The next step is to model
the hard information timeliness (HI T ) and soft information
timeliness (SI T ) fuzzy variables, whose respective crisp in-
puts are ITHand ITS. These two variables have the following
linguistic terms: HITRecent (trapezoidal, A= 0,B= 0,
C= 180,D= 540); HITO ld (trapezoidal, A= 360,
B= 540,C=,D=); SI TRecent (trapezoidal, A= 0,
B= 0,C= 500,D= 732) and SI TOld (trapezoidal,
A= 500,B= 732,C=,D=). OITHS is then
determined as the center of gravity of the fuzzified output of
the Mamdani fuzzy inference system (FIS) below:
IF HIT is H ITRecent AND SIT is SITRecent THEN
OITHS is OITRecent
is OITAcceptable
The set of membership functions associated with the
OITHS fuzzy variable are as follows:
OITOld :triangular (A= 0, B = 0, C = 0.5)
OITAcceptable :triangular (A= 0, B = 0.5, C = 1)
OITRecent :tiangular (A= 0.5, B = 1, C = 1)
In the case where only hard sources are available, OITH S
is computed by fuzzifying ITHvia a triangular membership
function with parameters A= 0,B= 180 and C= 540.
Finally, the robustness Rof the system is set to 1 since the
system can cope with real-time variations [21] [22].
In this Section, we empirically validate the proposed risk-
aware HSDF framework described in Sections III and IV. To
evaluate and contrast the effectiveness of the RMF with hard-
soft HLIF, we leaned on two scenarios: the east Canada coast
(ECC) outlined in [6] and a new one modeled around the Horn
of Africa (HOA).
A. Maritime Incident Reports
Fig. 2 reveals the distribution of worldwide maritime inci-
dents per vessel category in the POI under discussion. Trans-
port/utility and cargo vessels are nearly equally the subject
of maritime lawlessness given their strategic importance. As
expected, incidents related to military vessels are scant due to
their defensive capabilities.
The distribution of worldwide maritime incidents per in-
cident category is reflected in Fig. 3. Vessel boarding and
tresspassing (like in piracy cases) are responsible for 40%
of the criminal activity reported in the POI. The combined
effect of theft and crew damage accounts for almost the same
degree of unlawfulness. These facts underscore the importance
of automated HLIF solutions for a more effective MDA.
Industrial 29%
Transport or Utility
Fig. 2. Maritime incident distribution per vessel category
Crew Damage
Invasion 40%
17% Near Invasion
5% Threatened
3% Approach
Fig. 3. Maritime incident distribution per incident category
B. ECC Scenario Analysis
The ECC scenario consists of 14 maritime assets. An
extension of this scenario was tested successfully in [6]. The
Canadian maritime shoreline is clearly characterized by low
regional hostility as displayed in Fig. 4.
Fig. 4. Overall risk for all vessels in the ECC scenario (2 coast guard vessels,
2 medical vessels, 4 tugs, 1 oil tanker, 4 speed boats and 1 cruise)
ECCH1 0.163 0.600 0.833 0.0815
ECCHS (n= 10) 1 0.164 0.640 0.833 0.0874
ECCHS (n= 9) 1 0.164 0.640 0.833 0.0874
ECCHS (n= 8) 1 0.164 0.640 0.833 0.0874
ECCHS (n= 7) 1 0.164 0.640 0.833 0.0874
ECCHS (n= 6) 1 0.164 0.640 0.833 0.0874
ECCHS (n= 5) 1 0.164 0.640 0.833 0.0874
The overall effectiveness of the hard fusion scenario
(ECCH), as calculated by the metric in Section V, was 8.15%.
For the hard-soft scenario (ECCHS ) and despite the lack
of reported maritime incidents in the Canadian coastline,
the I F E value is slightly superior, around 8.74%. This is
caused by the minor increase in the IC indicator due to the
ingestion of a fairly reliable soft data source. Notice that
OITH=OITHS given that no compiled incident reports
apply to the ECC dataset. In this case, varying the number
of closest reports ndid not impact the RAL indicator at all.
C. HOA Scenario Analysis
The coast off the Horn of Africa is notorious for piracy,
as confirmed by the data extracted from the NGA WWTTS
reports. As seen below in Fig. 5, the density of maritime
incidents in this region is staggering.
Fig. 5. Horn of Africa maritime incidents
We created a simulated scenario of this volatile trade route
in order to test the HSDF capabilities of the RMF. The scenario
portrayed in Fig. 6 consists of 43 maritime units.
Fig. 6. Horn of Africa maritime scenario (4 medical vessels, 14 tugs, 11 oil
tankers, 12 speed boats and 2 cruises)
In this scenario, the risk evaluation process clearly benefited
from the incorporation of soft data. In Fig. 7, we find that the
regional hostility metric and risk of attack (as part of degree of
distress) soft risk features have a visible effect on the overall
risk of vessels.
For the HOAHS scenario, we find that the proposed HSDF
system improves on the HOAHscenario by around 16.89%
19.17%, as shown in Table V. The IBR value of 0.983 in the
HOAH1 0.163 0.600 0.833 0.0816
HOAHS (n= 10) 0.983 0.524 0.640 0.826 0.2723
HOAHS (n= 9) 0.983 0.526 0.640 0.826 0.2733
HOAHS (n= 8) 0.983 0.514 0.640 0.826 0.2671
HOAHS (n= 7) 0.983 0.511 0.640 0.826 0.2655
HOAHS (n= 6) 0.983 0.485 0.640 0.826 0.2520
HOAHS (n= 5) 0.983 0.482 0.640 0.826 0.2505
Fig. 7. HOA risk assessment. Left: hard scenario. Right: hard-soft scenario
hard-soft case indicates that a few vessels were still far away
from the nearest reported incidents. Furthermore, the RAL
of the HOAHS scenario experiences a drastic improvement
over its hard-only counterpart. This significant increase in
risk awareness is both accurate and useful for navigation in
such conflict-laden region. The downside of bringing soft data
into the fusion framework is expressed in the OIT indicator,
owing to the age of the available textual incidents (1.3 years
preceding the actual scenario date).
Overall, the HOA scenario provides useful information that
stimulates discussion on the cost/benefit of HSDF systems.
In our case study, we realized that the amalgamation of hard
and soft data can produce significant leaps in terms of IG.
Another lesson learned is that HSDF systems may not exhibit
the same IQ as hard fusion systems, as reliability/timeliness
may decline when ingesting soft sources.
In this work, we have augmented the RMF proposed in
[5] and [6] with an HSDF module as part of the risk feature
extraction and modeling. We tested our scheme with two MDA
scenarios where the number of maritime incident reports varies
substantially. In either case, our HSDF system has proven
effective in accurately translating hard and soft information
into a quantitative structure for risk analysis that correlates
with and confirms human intuition. Further still, our I F E
measure has proven to accurately translate important quali-
tative factors into relevant quantitative terms. This measure
could be applied to any data fusion system that deals with
risk. Even risk-agnostic systems could profit from the building
blocks behind the metric formalization provided that the
necessary customizations to the domain of interest are made.
Future work is concerned with HSDF in the response space
(Level 3 RMF, see Fig. 1) and the study of the dynamic nature
of the reliability of the data sources and their information.
[1] R. Abielmona, “Tackling big data in maritime domain awareness,
Vanguard, pp. 42–43, August-September 2013.
[2] E. Blasch, I. Kadar, J. Salerno, M. M. Kokar, G. M. Powell, D. D.
Corkill, and E. H. Ruspini, “Issues and challenges in situation assess-
ment (level 2 fusion),Journal of Advances in Information Fusion,
vol. 1, pp. 122–139, December 2006.
[3] E. Blasch and S. Plano, “DFIG level 5 (user refinement) issues sup-
porting situational assessment reasoning,” in International Conference
on Information Fusion, July 2005.
[4] A. H. Razavi, D. Inkpen, R. Falcon, and R. Abielmona, “Textual risk
mining for maritime situational awareness,” in 2014 IEEE International
Inter-Disciplinary Conference on Cognitive Methods in Situation Aware-
ness and Decision Support (CogSIMA), pp. 167–173, March 2014.
[5] R. Falcon, R. Abielmona, and A. Nayak, “An evolving risk management
framework for wireless sensor networks,” in Proceedings of the 2011
IEEE Int’l Conference on Computational Intelligence for Measurement
Systems and Applications (CIMSA), pp. 1–6, September 2011.
[6] R. Falcon and R. Abielmona, “A response-aware risk management
framework for search-and-rescue operations,” in 2012 IEEE Congress
on Evolutionary Computation (CEC), pp. 1540–1547, June 2012.
[7] ISO, “Risk management: Principles and guidelines,” International Or-
ganization for Standardization, no. 31000, 2009.
[8] X. Tan, Y. Zhang, X. Cui, and H. Xi, “Using hidden markov models
to evaluate the real-time risks of network,” in Proceedings of IEEE
Intl Symposium on Knowledge Acquisition and Modeling Workshop,
pp. 490–493, December 2008.
[9] K. Haslum and A. Arnes, “Real-time risk assessment using continuous-
time hidden markov models,” in Proceedings of Intl Conference on
Computational Intelligence and Security, pp. 1536–1540, November
[10] J. Merrick, J. van Dorp, and V. Dinesh, “Assessing uncertainty in
simulation based maritime risk assessment,” Risk Analysis, vol. 25,
pp. 731–743, July 2005.
[11] H. Abbass, A. Bender, S. Gaidow, and P. Whitbread, “Computational
red teaming: Past, present and future,” IEEE Computational Intelligence
Magazine, vol. 6, pp. 30–42, February 2011.
[12] A. Yang, H. Abbass, and R. Sarker, “Characterizing warfare in red
teaming,” IEEE Transactions on Systems, Man, and Cybernetics, Part
B: Cybernetics, vol. 36, no. 2, pp. 268–285, 2006.
[13] M. Jakob, O. Vanek, S. Urban, P. Benda, and M. Pechoucek, “Adversarial
modeling and reasoning in the maritime domain - first year report,” Agent
Technology Center, Department of Cybernetics, FEE Czech Technical
University in Prague, 2009.
[14] N. Friedman, The Naval Institute Guide to World Naval Weapon Systems.
Naval Institute Press, 2006.
[15] S. Dewey and J. Maynard, “Time, technology and exploitation can
future military command-and-control domination be linked to effective
knowledge?,Joint Military Operations Department Naval War College,
April 2008.
[16] I. Lim and F. Jau, “Comprehensive maritime domain awareness: an idea
whose time has come?,” Defence, Terrorism and Security, Globalisation
and International Trade, October 2007.
[17] K. Moore, “Predictive analysis for naval deployment activities,PANDA
BAA, no. 05-44, 2005.
[18] R. Laxhammar, “Artificial intelligence for situation assessment,” Mas-
ter’s thesis, Royal Institute of Technology, Sweden, 2007.
[19] E. Blasch, E. Dorion, P. Valin, E. Bosse, and J. Roy, “Ontology
alignment in geographical hard-soft information fusion systems,” in 13th
Conference on Information Fusion (FUSION), pp. 26–29, July 2010.
[20] H. Shao, R. Abielmona, R. Falcon, and N. Japkowicz, “Vessel track
correlation and association using fuzzy logic and echo state networks,”
in IEEE Congress on Evolutionary Computation (CEC), to appear, 2014.
[21] E. Blasch, P. Valin, and E. Bosse, “Measures of effectiveness for high-
level fusion,” in 13th Conference on Information Fusion (FUSION),
pp. 1–8, July 2010.
[22] E. Blasch, R. Breton, and P. Valin, “Information fusion measures of
effectiveness (MOE) for decision support,Proc. of SPIE, vol. 8050,
pp. 805011–1–805011–12, May 2011.
... These parameter values make the MIP metric fairly sensitive to the presence of maritime incidents. • Mean Incident Severity (MIS): Following [22], MIS is calculated as the mean severity of the n nearest incidents within max distance of the vessel. The incident severities are given in Table 1 of [22]. ...
... • Mean Incident Severity (MIS): Following [22], MIS is calculated as the mean severity of the n nearest incidents within max distance of the vessel. The incident severities are given in Table 1 of [22]. • Vessel Pertinence Index (VPI): As in [22], VPI is the maximum relevance of the n nearest incidents within max distance of the vessel. ...
... The incident severities are given in Table 1 of [22]. • Vessel Pertinence Index (VPI): As in [22], VPI is the maximum relevance of the n nearest incidents within max distance of the vessel. The similarities of the vessel categories is given in Table 2 of [22]. ...
Conference Paper
Full-text available
Assessing the overall risk of maritime entities over time is a pivotal issue in Maritime Risk Assessment (MRA). This process largely hinges on human expertise to characterize the risk features of interest as well as their mathematical underpinnings and the way in which their local risk outputs are amalgamated to produce an overall risk value. In this paper, we take steps towards automating this time-consuming endeavour by learning the fuzzy rule base (FRB) governing the Risk Assessment module of an existing Risk Management Framework (RMF) directly from data. We employ several genetic fuzzy systems (GFSs) available in KEEL to optimize the membership functions and learn the structure of a FRB used for risk assessment purposes. The proposed methodology is illustrated with two case studies in maritime surveillance. The empirical results indicate that the New Slave (NSLV) algorithm generated the most accurate FRBs, while the Hybrid Genetics-Based Machine Learning (GBML) algorithm generated the most interpretable FRBs. To the best of our knowledge, this is the first attempt to apply GFSs in MRA.
... Falcon, Abielmona, Billings, et al. 23 explore MoEs that are useful in a risk management framework (RMF) architecture that characterizes and evaluates maritime vessels in potential distress. An extension to ITS is suggested in the last paragraph of this section. ...
... Falcon, Abielmona, Billings, et al. 23 utilize the general information fusion effectiveness (IFE) MoE developed by Blasch, Valin, and Bosse 22 that contains three multiplicative factors: information gain, information quality, 210 Chapter 6 ...
... Robustness is set to 1 since the system can function with real-time variations. 23 This IFE MoE can be extended to ITS in a variety of ways: by considering vehicles rather than vessels in the architecture; using studies to evaluate the IBR as a function of vehicle location (by geographical location in a country, state, county, or city; by roadway type and by lane-Interstate, urban, rural, unimproved; by congestion level; etc.) and the types or population of vehicles in use at a particular time; evaluating drivers' RAL, perhaps with surveys; investigating the reliability of ROI and ROS as they pertain to the various sources of traffic data described in Section 2.3; adapting the IT H and IT S concepts by considering the delays attributable to the hard and soft traffic data sources, respectively, as they affect drivers, other travelers (e.g., transit users), traffic management personal, and other stakeholders; and incorporating additional suitable analogies. ...
There are ongoing needs for data fusion research in many areas, including: • Reliability and quality of input data to the fusion system. • Security of the data fusion system. • Fusion of hard and soft data. • Assessing the fusion system using measures of performance (MoPs). • Adaptive nature of MoPs. • MoP dependence on operational scenario. • Measures of effectiveness (MoEs) for information fusion. • Ground truth. • Commercial database management and operating systems suitability. • Designing for worst-case data transmission and processing scenarios.
... structured, quantitative, more objective, usually sensed data) and soft data (i.e. unstructured/semi-structured, more subjective, qualitative data, such as textual reports on vessel sightings or marine incidents) collected in different maritime sectors as shown in Table 1 (Falcon et al. 2014). Different maritime organizations are making efforts to exploit these data and attempt to distill constructive information to guide their daily operations and decision making. ...
... Most literature in maritime big data focuses on the Maritime Domain Awareness (MDA) and intelligent ship navigation. MDA refers to the understanding of regional physical and environmental maritime conditions (Falcon et al. 2014). Big data applications have been studied in different MDA processes and behaviors including maritime situational awareness (Cazzanti and Pallotta 2015;Shahir et al. 2015), anomaly detection (Pallotta and Jousselme 2015;Shahir et al. 2015), trajectory prediction (Plachkov et al. 2016), intent assessment and threat assessment (Brüggemann et al. 2016;Santipantakis, Kotis, and Vouros 2015). ...
Full-text available
50 free online copies of the full text Big data analytics has prospered in recent years and has triggered revolutionary changes in various industries. However, its adoption in maritime organizations is relatively lagged and there is no study addressing this phenomenon so far. This paper develops a fuzzy Delphi-AHP-TOPSIS framework to identify barriers in emerging technology adoption. A case study employs this framework to investigate the hurdles for big data analytics to be adopted in the maritime industry. The case study shows that the framework is useful in identifying and ranking barriers. The results unveil the most serious managerial, cultural, and technical barriers that impede the adoption of big data analytics in maritime organizations.
... The authors of [12] proposed a risk-aware fusion effectiveness metric of hard and soft data fusion in maritime domain awareness. Given that the main purpose of our work is also to predict the quality of fused images, the concepts of the risk-aware fusion effectiveness metric can be combined with the GIQE to predict the QoI of both fused and non-fused images. ...
... When the commonly used AIS data to track vessels is unavailable, for example, when vessels turn off their transponders to "go dark", allowing them to hide illicit activity such as illegal fishing or modern slavery, or because some nations, including Canada, do not require fishing vessels or small boat owners to use AIS at all, one can resort to remote sensing data to light them back up [2]. However, the satellite coverage can be limited due to either non-overlapping sensor footprints or on-board sensor resolution [3] hence it is expected that vessel detection and situational awareness can be improved by using This Prior works have highlighted [4] or reported [5] the benefits of hard and soft data fusion specifically in the maritime domain. In other application domains, the benefits are reported in [6] where the modified Dempster's rule of combination for prediction of social events is used, in [7] where argumentation is integrated with the fusion of hard and soft data in application to threat assessment in multiple domains, while in [8] a method for multi-level fusion of hard and soft information for intelligence is presented. ...
... The Automatic Identification System (AIS) is an automatic tracking system which provides rich information such as Maritime Mobile Service Identify (MMSI), longitude, latitude, speed over ground, destination and so on. As an emerging technology to make ships "visible" to each other, AIS is mainly used for collision avoidance [1][2][3] and ship traffic management system [4][5][6]. In this paper, we make the first attempt for extracting ship routes by using AIS data, and then conduct a study on ship route clustering analysis. ...
... The traditional statistic method not only waste manpower and material but also may be not accurate. The Automatic Identification System (AIS) is an automatic tracking system that mainly used for collision avoidance [1][2][3] and ship traffic management system [4][5][6]. It provides rich information such as Maritime Mobile Service Identify (MMSI), longitude, latitude, speed over ground, destination and so on. ...
This chapter uses genetic fuzzy systems (GFS) to assess the risk level of maritime vessels transmitting Automatic Identification System (AIS) data. Previous risk assessment approaches based on fuzzy inference systems (FIS) relied on domain experts to specify the FIS membership functions as well as the fuzzy rule base (FRB), a burdensome and time-consuming process. This chapter aims to alleviate this burden by learning the membership functions and FRB for the FIS of an existing Risk Management Framework (RMF) directly from data. The proposed methodology is tested with four different case studies in maritime risk analysis. Each case study concerns a unique scenario involving a particular region: the Gulf of Guinea, the Strait of Malacca, the Northern Atlantic during a storm, and the Northern Atlantic during a period of calm seas. The experiments compare 14 GFS algorithms from the KEEL software package and evaluate the resulting FRBs according to their accuracy and interpretability. The results indicate that IVTURS, LogitBoost, and NSLV generate the most accurate rule bases while SGERD, GCCL, NSLV, and GBML each generate interpretable rule bases. Finally, IVTURS, NSLV, and GBML algorithms offer a reasonable compromise between accuracy and interpretability.
Wireless sensor networks consist of several tiny low cost sensor nodes that are deployed for many applications such as military, civil, industrial, healthcare, home automation, etc. Recent technological developments have enabled the use of wireless visual sensor networks (WVSNs) for sensitive applications such as video surveillance and monitoring applications. Limited memory, energy and bandwidth are the major constraints in WVSN that can be simplified by the use of compressed sensing (CS), which asserts that sparse signals can be reconstructed from very few measurements. CS a computational intelligence solution is about acquiring and recovering the signal in the most efficient manner possible using incoherent projection basis. In the case of video surveillance applications, the entire video may not be useful hence, with the help of efficient algorithms the presence of the anomalies can be detected and transmitted to help user at the monitoring site to take necessary action. In this chapter, particle swarm optimization (PSO) based efficient anomaly detection system (EADS) is proposed which will detect the presence of anomalies and transmit the required measurements via TelosB nodes to the network operator. This system adopts the concept of CS to obtain the compressive measurements so that the object detection algorithm can be applied to the measurements rather than samples. PSO is employed for optimizing compressive measurements while a mean based measurement differencing approach is used for detecting the object. This proposed efficient system has the intelligence of detecting targets with fewer measurements and transmit the required compressive measurements for reconstruction with less energy, thereby increasing the network lifetime. PSO is used to optimize the transmission distance with minimum number of hops towards destination, to achieve reduced energy consumption. However, the lifetime of the network is still bounded by batteries, the sole source of energy in WVSNs. Alternative energy utilization can be effectively included to recharge the batteries on-board and extend the lifetime of the network. Solar energy harvesting forms an effective resource due to its ambient presence. Hence, solar energy harvester is incorporated in the proposed EADS to extend its lifetime.
Conference Paper
Rough Cognitive Ensembles (RCEs) has recently emerged as a granular multiclassifier system composed of a set of Rough Cognitive Networks (RCNs), each operating at a different granularity degree. While this model is capable of outperforming several traditional classifiers reported in the literature, there is still room for enhancing its performance. In this paper, we propose a fuzzy strategy to activate the RCN input neurons before performing the inference process. This fuzzy activation mechanism quantifies the extent to which an object belongs to the intersection between its similarity class and each granular region in the RCN topology. The numerical simulations have shown that the improved ensemble classifier significantly outperforms the original RCE model for the adopted datasets. After comparing the proposed model to 14 well-known classifiers, the experimental evidence confirms that our scheme yields improved classification rates.
Conference Paper
Full-text available
Tracking moving objects is a task of the utmost importance to the defence community. As this task requires high accuracy, rather than employing a single detector, it has become common to use multiple ones. In such cases, the tracks produced by these detectors need to be correlated (if they belong to the same sensing modality) or associated (if they were produced by different sensing modalities). In this work, we introduce Computational-Intelligence-based methods for correlating and associating various contacts and tracks pertaining to maritime vessels in an area of interest. Fuzzy k-Nearest Neighbors will be used to conduct track correlation and Fuzzy C-Means clustering will be applied for association. In that way, the uncertainty of the track correlation and association is handled through fuzzy logic. To better model the state of the moving target, the traditional Kalman Filter will be extended using an Echo State Network. Experimental results on five different types of sensing systems will be discussed to justify the choices made in the development of our approach. In particular, we will demonstrate the judiciousness of using Fuzzy k-Nearest Neighbors and Fuzzy C-Means on our tracking system and show how the extension of the traditional Kalman Filter by a recurrent neural network is superior to its extension by other methods.
Full-text available
This article examines ISO 31000-2009, also referred to as ANSI/ASSE Z690.2, the second of a trio of standards dealing with the concept of Risk. In our first review (Luko 20136. Luko , S. N. ( 2013 ). Risk management terminology . Quality Engineering , 25 ( 3 ): 292 – 297 . [Taylor & Francis Online], [Web of Science ®]View all references) risk management terminology was reviewed. The terminology documents, ISO Guide 73 and ANSI/ASSE Z690.1-2011, were found to be identical and contained all of the risk vocabulary used in the subsequent two standards. In the present review, the second of the trio of standards concerning risk is treated. The general topic of this standard is risk management principles and guidelines.
Conference Paper
Full-text available
In this paper, we propose an auxiliary Machine Learning (ML) and Natural Language Processing (NLP) integrated system for maritime situational awareness (MSA) operations. We bring into account a new and influential asset – human intuition and perception – to the existing semi-automated decision support systems that mostly rely on numerical data collected by electronic sensors or cameras located either directly on the vessels or in the maritime command-and-control centers. For our project, we gathered weekly textual reports spanning twelve months from the United States Worldwide Threats to Shipping Reports repository that belongs to the National Geospatial-Intelligence Agency (NGA). We considered the maritime incident reports written by human operators as a valuable and accessible unstructured textual input source in which a span of text 1 is called "risk" if it expresses one of the following kinds of vessel incidents: fired, robbed, boarded, hijacked, attacked, chased, approached, kidnapped, boarding attempted, suspiciously approached or clashed with. Our approach benefits from probability distributions of some useful features annotated based on a list of lexicons that contain expressions denoting vessel types, risks types, risk associates, maritime geographical locations, dates and times. These distributions are captured and used to anchor the span of "risks" as they are described in the textual reports. After some pre-processing steps that include tokenization, named entity extraction and part-of-speech tagging, the textual risk mining system applies a variety of sequence classification algorithms, e.g., Conditional Random Fields, Conditional Markov Models and Hidden Markov Models in order to compare the risk classification performance. Empirical results show that our NLP/ML-based system can extract variable-length risk spans from the textual reports with about 90% correctness.
Full-text available
For decades, there have been discussions on measures of merits (MOM) that include measures of effectiveness (MOE) and measures of performance (MOP) for system-level performance. As the amount of sensed and collected data becomes increasingly large, there is a need to look at the architectures, metrics, and processes that provide the best methods for decision support systems. In this paper, we overview some information fusion methods in decision support and address the capability to measure the effects of the fusion products on user functions. The current standard Information Fusion model is the Data Fusion Information Group (DFIG) model that specifically addresses the needs of the user in an information fusion system. Decision support implies that information methods augment user decision making as opposed to the machine making the decision and displaying it to user. We develop a list of suggested measures of merits that facilitate decision support decision support Measures of Effectiveness (MOE) metrics of quality, information gain, and robustness, from the analysis based on the measures of performance (MOPs) of timeliness, accuracy, confidence, throughput, and cost. We demonstrate in an example with motion imagery to support the MOEs of quality (time/decision confidence plots), information gain (completeness of annotated imagery for situation awareness), and robustness through analysis of imagery over time and repeated looks for enhanced target identification confidence.
Conference Paper
Full-text available
Efficient coordination among all assets participating in a response to a search-and-rescue (SAR) incident has long been a focus of many governments and organizations. Finding inno-vative solutions that guarantee a swift reaction to the distressed entity with a rational use of the available resources is pivotal to the success of the SAR operation. In spite of the plethora of successfully deployed SAR systems, we witness a substantial gap when it comes to the integration of risk-driven analyses into the underlying machinery of any decision support platform that leans upon the in-field SAR assets. This paper extends a recently proposed risk management framework [1] by adding automated modules for risk monitoring and response selection. An evolutionary multi-objective optimization algorithm is used to navigate across the discrete space of all available assets and their set of actions in order to present a limited number of promising responses to a SAR operator, who will ultimately decide what action must be carried out. The proposed methodology was validated in the context of a simulated nautical SAR scenario in the Canadian Atlantic coastline with nine different types of ground, maritime and aerial assets.
Conference Paper
Full-text available
Individual units in a wireless sensor network (WSN) are exposed to multiple risks, either during or after their deployment. The identification of the risk sources and their watchful monitoring in dynamic, unpredictable environments is pivotal to ensure a smooth, long-term functioning of the WSN. We introduce an evolving risk management framework for WSNs that captures multiple risk features and provides both a visual depiction of the corporate network threats at any time and a numerical assessment of any sensor's overall risk. The visualization module is embodied through an evolving clustering architecture which heavily relies on shadowed sets. The risk assessment module embraces fuzzy and shadowed evaluations of the risk sources and incorporates a simple adaptive learning process that weights the risk sources proportionally to their observed impact on failed sensors. A distinctive trait of the proposed framework is its highly automated yet still human-centric nature. Experiments utilizing different sensor models and deployment scenarios confirm the feasibility of the risk management platform under consideration.
Highly effective employment of military force at the operational level will require greater levels of synchronization and harmonization to maximize economy of force and thus combat potential. The evolution of Operational Command and Control (C2) will remain, and become increasingly so, the primary enabler for U.S. joint military operations to accomplish this task. The most effective strategy to optimize this operational function is by increasing the speed of the operational commander's decision-making process by compressing the decision-making cycle. Factor Time must be manipulated and exploited. This paper seeks to determine if the emerging field of Knowledge Management (KM) can successfully apply a three-pronged approach (Knowledge Transfer Systems, Systems Shaping, and Complex Adaptive System Analysis) which can effect this needed change. While emphasizing human-centricity, KM's technological applications are simply tools created to increase human capacity. If this holistic approach is possible, and significant improvement is realized, it will be incumbent upon the commander to find ways to fully exploit this improved capability. As history has shown, superior C2 structures can achieve battle-space domination.
The principal features of distributed intelligent (DI) system include a diversity of perspectives arising from separated individuals and a problem-solving method whereby the individuals can coordinate their activities to accomplish a collective objective. These features require numerous capabilities and supporting structures. A system that embodies such capabilities, however, seems especially well suited to situation assessment (SA) tasks. These tasks require the collection, analysis, and interpretation of widely disparate data. Moreover, comparably complex and distributed systems are often required to disseminate the results of these analyses effectively. This paper explores the relationships among structural capabilities of DI systems and functional requirements of effective SA systems. Discussed are some of the options and trade-offs of various architectural decisions in the context of two contrasting DI designs. (Author)