Content uploaded by J E T Akinsola
Author content
All content in this area was uploaded by J E T Akinsola on Jul 11, 2017
Content may be subject to copyright.
Content uploaded by J E T Akinsola
Author content
All content in this area was uploaded by J E T Akinsola on Jul 11, 2017
Content may be subject to copyright.
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 128
Supervised Machine Learning Algorithms:
Classification and Comparison
Osisanwo F.Y.*1, Akinsola J.E.T.*2, Awodele O.*3, Hinmikaiye J. O.*4, Olakanmi O.*5,Akinjobi J.**6
*Department of Computer Science, Babcock University, Ilishan-Remo, Ogun State, Nigeria.
**Department of Computer Science, Crawford University, Igbesa, Ogun State, Nigeria
Abstract ---- Supervised Machine Learning (SML) is
the search for algorithms that reason from
externally supplied instances to produce general
hypotheses, which then make predictions about
future instances. Supervised classification is one of
the tasks most frequently carried out by the
intelligent systems. This paper describes various
Supervised Machine Learning (ML) classification
techniques, compares various supervised learning
algorithms as well as determines the most efficient
classification algorithm based on the data set, the
number of instances and variables (features).Seven
different machine learning algorithms were
considered:Decision Table, Random Forest (RF) ,
Naïve Bayes (NB) , Support Vector Machine (SVM),
Neural Networks (Perceptron), JRip and Decision
Tree (J48) using Waikato Environment for
Knowledge Analysis (WEKA)machine learning
tool.To implement the algorithms, Diabetes data set
was used for the classification with 786 instances
with eight attributes as independent variable and
one as dependent variable for the analysis. The
results show that SVMwas found to be the algorithm
with most precision and accuracy. Naïve Bayes and
Random Forest classification algorithms were found
to be the next accurate after SVM accordingly. The
research shows that time taken to build a model and
precision (accuracy) is a factor on one hand; while
kappa statistic and Mean Absolute Error (MAE) is
another factor on the other hand. Therefore, ML
algorithms requires precision, accuracy and
minimum error to have supervised predictive
machine learning.
Keywords: Machine Learning, Classifiers, Data
Mining Techniques, Data Analysis, Learning
Algorithms, Supervised Machine Learning
INTRODUCTION
Machine learning is one of the fastest growing
areas of computer science, with far-reaching
applications. It refers to the automated detection of
meaningful patterns in data. Machine learning tools
are concerned with endowing programs with the
ability to learn and adapt [19].
Machine Learning has become one of the mainstays
of Information Technology and with that, a rather
central, albeit usually hidden, part of our life. With
the ever increasing amounts of data becoming
available there is a good reason to believe that smart
data analysis will become even more pervasive as a
necessary ingredient for technological progress.
There are several applications for Machine
Learning (ML), the most significant of which is data
mining. People are often prone to making mistakes
during analyses or, possibly, when trying to
establish relationships between multiple features [9].
Data Mining and Machine Learning are
Siamese twins from which several insights can be
derived through proper learning algorithms. There
has been tremendous progress in data mining and
machine learning as a result of evolution of smart
and Nano technology which brought about curiosity
in finding hidden patterns in data to derive value.
The fusion of statistics, machine learning,
information theory, and computing has created a
solid science, with a firm mathematical base, and
with very powerful tools.
Machine learning algorithms are organized into
a taxonomy based on the desired outcome of the
algorithm. Supervised learning generates a function
that maps inputs to desired outputs.
Unprecedented data generation has made
machine learning techniques become sophisticated
from time to time. This has called for utilization for
several algorithms for both supervised and
unsupervised machine learning. Supervised learning
is fairly common in classification problems because
the goal is often to get the computer to learn a
classification system that we have created [21].
ML is perfectly intended for accomplishing the
accessibility hidden within Big Data. ML hand
over’s on the guarantee of extracting importance
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 129
from big and distinct data sources through outlying
less dependence scheduled on individual track as it
is data determined and spurts at machine scale.
Machine learning is fine suitable towards the
intricacy of handling through dissimilar data origin
and the vast range of variables as well as amount of
data concerned where ML prospers on increasing
datasets. The extra data supply into a ML structure,
the more it be able to be trained and concern the
consequences to superior value of insights. At the
liberty from the confines of individual level thought
and study, ML is clever to find out and show the
patterns hidden in the data [15].
One standard formulation of the supervised
learning task is the classification problem: The
learner is required to learn (to approximate the
behavior of) a function which maps a vector into
one of several classes by looking at several input-
output examples of the function. Inductive machine
learning is the process of learning a set of rules from
instances (examples in a training set), or more
generally speaking, creating a classifier that can be
used to generalize from new instances. The process
of applying supervised ML to a real-world problem
is described in Figure 1.
Figure 1: The Processes of Supervised Machine
Learning
This work focuses on the classification of ML
algorithms and determining the most efficient
algorithm with highest accuracy and precision. As
well as establishing the performance of different
algorithms on large and smaller data sets with a
view classify them correctly and give insight on
how to build supervised machine learning models.
The remaining part of this work is arranged as
follows: Section 2 presents the literature review
discussing classification of different supervised
learning algorithms; section 3 presents the
methodology used, section 4 discusses the results of
the work while section 5 gives the conclusion and
recommendation for further works.
I. LITERATURE REVIEW
A. Classification of Supervised Learning
Algorithms
According to [21], the supervised machine
learning algorithms which deals more with
classification includes the following: Linear
Classifiers, Logistic Regression, Naïve Bayes
Classifier, Perceptron, Support Vector Machine;
Quadratic Classifiers, K-Means Clustering,
Boosting, Decision Tree, Random Forest (RF);
Neural networks, Bayesian Networks and so on.
1) Linear Classifiers: Linear models for
classification separate input vectors into classes
using linear (hyperplane) decision boundaries [6].
The goal of classification in linear classifiers in
machine learning, is to group items that have similar
feature values, into groups. [23] stated that a linear
classifier achieves this goal by making a
classification decision based on the value of the
linear combination of the features. A linear classifier
is often used in situations where the speed of
classification is an issue, since it is rated the fastest
classifier [21].Also, linear classifiers often work
very well when the number of dimensions is large,
as in document classification, where each element is
typically the number of counts of a word in a
document. The rate of convergence among data set
variables however depends on the margin. Roughly
speaking, the margin quantifies how linearly
separable a dataset is, and hence how easy it is to
solve a given classification problem [18].
2) Logistic regression: This is a classification
function that uses class for building and uses a
single multinomial logistic regression model with a
single estimator. Logistic regression usually states
where the boundary between the classes exists, also
states the class probabilities depend on distance
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 130
from the boundary, in a specific approach. This
moves towards the extremes (0 and 1) more rapidly
when data set is larger. These statements about
probabilities which make logistic regression more
than just a classifier. It makes stronger, more
detailed predictions, and can be fit in a different
way; but those strong predictions could be wrong.
Logistic regression is an approach to prediction, like
Ordinary Least Squares (OLS) regression. However,
with logistic regression, prediction results in a
dichotomous outcome [13]. Logistic regression is
one of the most commonly used tools for applied
statistics and discrete data analysis. Logistic
regression is linear interpolation[11].
3) Naive Bayesian (NB) Networks: These
are very simple Bayesian networks which are
composed of directed acyclic graphs with only one
parent (representing the unobserved node) and
several children (corresponding to observed nodes)
with a strong assumption of independence among
child nodes in the context of their parent [7].Thus,
the independence model (Naive Bayes) is based on
estimating [14]. Bayes classifiers are usually less
accurate that other more sophisticated learning
algorithms (such as ANNs).However, [5] performed
a large-scale comparison of the naive Bayes
classifier with state-of-the-art algorithms for
decision tree induction, instance-based learning, and
rule induction on standard benchmark datasets, and
found it to be sometimes superior to the other
learning schemes, even on datasets with substantial
feature dependencies. Bayes classifier has attribute-
independence problem which was addressed with
Averaged One-Dependence Estimators [8].
4) Multi-layer Perceptron: This is a
classifier in which the weights of the network are
found by solving a quadratic programming problem
with linear constraints, rather than by solving a non-
convex, unconstrained minimization problem as in
standard neural network training [21].Other well-
known algorithms are based on the notion of
perceptron [17].Perceptron algorithm is used for
learning from a batch of training instances by
running the algorithm repeatedly through the
training set until it finds a prediction vector which is
correct on all of the training set. This prediction rule
is then used for predicting the labels on the test set
[9].
5) Support Vector Machines (SVMs): These
are the most recent supervised machine learning
technique [24].Support Vector Machine (SVM)
models are closelyrelated to classical multilayer
perceptron neural networks.SVMs revolve around
the notion of a ―margin‖—either side of a
hyperplane that separates two data classes.
Maximizing the margin and thereby creating the
largest possible distance between the separating
hyperplane and the instances on either side of it has
been proven to reduce an upper bound on the
expected generalisation error [9].
6) K-means: According to [2] and [22]K-
means is one of the simplest unsupervised learning
algorithms that solve the well-known clustering
problem. The procedure follows a simple and easy
way to classify a given data set through a certain
number of clusters (assume k clusters) fixed a
priori.K-Means algorithm is be employed when
labeled data is not available [1].General method of
converting rough rules of thumb into highly accurate
prediction rule. Given ―weak‖ learning algorithm
that can consistently find classifiers (―rules of
thumb‖) at least slightly better than random, say,
accuracy _ 55%, with sufficient data, a boosting
algorithm can provably construct single classifier
with very high accuracy, say, 99% [16].
7) Decision Trees: Decision Trees (DT) are
trees that classify instances by sorting them based
on feature values. Each node in a decision tree
represents a feature in an instance to be classified,
and each branch represents a value that the node can
assume. Instances are classified starting at the root
node and sorted based on their feature values
[9].Decision tree learning, used in data mining and
machine learning, uses a decision tree as a
predictive model which maps observations about an
item to conclusions about the item's target value.
More descriptive names for such tree models are
classification trees or regression trees [20].Decision
tree classifiers usually employ post-pruning
techniques that evaluate the performance of decision
trees, as they are pruned by using a validation set.
Any node can be removed and assigned the most
common class of the training instances that are
sorted to it [9].
8) Neural Networks:[2]opined Neural
Networks (NN) that can actually perform a number
of regression and/or classification tasks at once,
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 131
although commonly each network performs only
one. In the vast majority of cases, therefore, the
network will have a single output variable, although
in the case of many-state classification problems,
this may correspond to a number of output units (the
post-processing stage takes care of the mapping
from output units to output variables).Artificial
Neural Network (ANN) depends upon three
fundamental aspects, input and activation functions
of the unit, network architecture and the weight of
each input connection. Given that the first two
aspects are fixed, the behavior of the ANN is
defined by the current values of the weights. The
weights of the net to be trained are initially set to
random values, and then instances of the training set
are repeatedly exposed to the net. The values for the
input of an instance are placed on the input units and
the output of the net is compared with the desired
output for this instance. Then, all the weights in the
net are adjusted slightly in the direction that would
bring the output values of the net closer to the
values for the desired output. There are several
algorithms with which a network can be trained
[12].
9) Bayesian Network: A Bayesian Network
(BN) is a graphical model for probability
relationships among a set of variables (features).
Bayesian networks are the most well-known
representative of statistical learning algorithms
[9].The most interesting feature of BNs, compared
to decision trees or neural networks, is most
certainly the possibility of taking into account prior
information about a given problem, in terms of
structural relationships among its features [9].A
problem of BN classifiers is that they are not
suitable for datasets with many features [4].This
prior expertise, or domain knowledge, about the
structure of a Bayesian network can take the
following forms:
1. Declaring that a node is a root node, i.e., it
has no parents.
2. Declaring that a node is a leaf node, i.e., it
has no children.
3. Declaring that a node is a direct cause or
direct effect of another node.
4. Declaring that a node is not directly
connected to another node.
5. Declaring that two nodes are independent,
given a condition-set.
6. Providing partial nodes ordering, that is,
declare that a node appears earlier than
another node in the ordering.
7. Providing a complete node ordering.
B. Features of Machine Learning Algorithms
Supervised machine learning techniques are
applicable in numerous domains. A number of
Machine Learning (ML) application oriented papers
can be found in [18], [25].
Generally, SVMs and neural networks tend to
perform much better when dealing with multi-
dimensions and continuous features. On the other
hand, logic-based systems tend to perform better
when dealing with discrete/categorical features. For
neural network models and SVMs, a large sample
size is required in order to achieve its maximum
prediction accuracy whereas NB may need a
relatively small dataset.
There is general agreement that k-NN is very
sensitive to irrelevant features: this characteristic
can be explained by the way the algorithm works.
Moreover, the presence of irrelevant features can
make neural network training very inefficient, even
impractical.Most decision tree algorithms cannot
perform well with problems that require diagonal
partitioning. The division of the instance space is
orthogonal to the axis of one variable and parallel to
all other axes. Therefore, the resulting regions after
partitioning are all hyperrectangles. The ANNs and
the SVMs perform well when multi-collinearity is
present and a nonlinear relationship exists between
the input and output features.
Naive Bayes (NB) requires little storage space
during both the training and classification stages: the
strict minimum is the memory needed to store the
prior and conditional probabilities. The basic kNN
algorithm uses a great deal of storage space for the
training phase, and its execution space is at least as
big as its training space. On the contrary, for all
non-lazy learners, execution space is usually much
smaller than training space, since the resulting
classifier is usually a highly condensed summary of
the data. Moreover, Naive Bayes and the kNN can
be easily used as incremental learners whereas rule
algorithms cannot. Naive Bayes is naturally robust
to missing values since these are simply ignored in
computing probabilities and hence have no impact
on the final decision. On the contrary, kNN and
neural networks require complete records to do their
work.
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 132
Finally, Decision Trees and NB generally have
different operational profiles, when one is very
accurate the other is not and vice versa. On the
contrary, decision trees and rule classifiers have a
similar operational profile. SVM and ANN have
also a similar operational profile. No single learning
algorithm can uniformly outperform other
algorithms over all datasets.
Different data sets with different kind of
variables and the number of instances determine the
type of algorithm that will perform well. There is no
single learning algorithm that will outperform other
algorithms based on all data sets according to no
free lunch theorem. [10] Table 1 presents the
comparative analysis of various learning algorithms.
Table 1: Comparing learning algorithms (**** stars represent the best and * star the worst performance)[9]
II. RESEARCH METHODOLOGY
The research data was obtained from National
Institute of Diabetes and Digestive and Kidney
Diseases which
was made available online at University of
California,
Irvive website:
https://archive.ics.uci.edu/ml/machine-learning
databases/pima-indians-diabetes/ (2017). This data
was chosen because of its accuracy and has also
been anonymized (de-identifed), therefore
confidentiality is ensured. The number of Attributes
is 8 with one class making it 9. All attributes are
numeric-valued as follows:
1. Number of times pregnant
2. Plasma glucose concentration a 2 hours in an
oral glucose tolerance test
3. Diastolic blood pressure (mm Hg)
4. Triceps skin fold thickness (mm)
5. 2-Hour serum insulin (mu U/ml)
6. Body mass index (weight in kg/(height in m)^2)
7. Diabetes pedigree function
8. Age (years)
9. Class variable (0 or 1)
Table 2: Class Distribution: (class value 1 is
interpreted as "tested positive for diabetes") and
(class value 0 is interpreted as "tested negative for
diabetes")
Class
Value Number of
instances
Converted Value
(attribute)
0
500
NO
1
268
YES
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 133
Table 2 shows 768 as the total number of instances
used for this research work with 500 tested positive
for diabetes and 268 tested negative for diabetes.
The comparative analysis among various
supervised machine learning algorithms was carried
out using WEKA 3.7.13 (WEKA - Waikato
Environment for Knowledge Analysis). The data set
was trained to reflect one nominal attribute column
as the dependent variable. The values 1’s for class
distribution (class variable) were changed to YES
which means tested POSITIVE for DIABETS and
values 0s for class distribution (class variable) were
changed NO which means tested NEGATIVE for
DIABETES. This is essential because most of the
algorithms require that there must be at least one
nominal variable column. Seven classification
algorithms were used in the course of this research
namely: Decision Table, Random Forest, Naïve
Bayes, SVM, Neural Networks (Perceptron), JRip
and Decision Tree (J48). The following attributes
were considered for the comparative analysis: Time,
Correctly Classified, Incorrectly Classified, Test
Mode, No of instances, Kappa statistic, MAE,
Precision of YES, Precision of NO and
Classification.
In order to predict the accuracy and ensure precision
for different machine learning algorithms, this
research work was carried out by tuning the
parameters with two different sets of number of
instances. The first category was 768 instances and
9 attributes as follows (Number of times pregnant,
Plasma glucose concentration a 2 hours in an oral
glucose tolerance test, Diastolic blood pressure (mm
Hg), Triceps skin fold thickness (mm), 2-Hour
serum insulin (mu U/ml), Body mass index (weight
in kg/(height in m)^2), Diabetes pedigree function,
Age (years) and Class variable (0 or 1)) with one
dependent variable and eight independent variables.
The second category of data set was 384 instances
and 6 attributes as follows (Number of times
pregnant, Plasma glucose concentration a 2 hours in
an oral glucose tolerance test, 2-Hour serum insulin
(mu U/ml), Diabetes pedigree function, Age (years)
and Class variable (0 or 1)) with one dependent
variable and five independent variables.
IV. RESULTS AND DISCUSSION
A. Results
WEKA was used in the classification and
comparison of the various machine leaning
algorithms. Table 3 shows the resultswith 9
attributes as well as parameters considered.
Table 3: Comparison of various classification algorithms with large data set and more attributes
Algorithm
Time
(Sec)
Correctly
Classified
(%)
Incorrectly
Classified
(%)
Test
Mode
Attributes
No of
instances
Kappa
statistic
MAE
Precision
of YES
Precision
of NO
Classifi-
cation
Decision
Table
0.23
72.3958
27.6042
10-fold
cross-
validation
9
768
0.3752
0.341
0.619
0.771
Rules
Random
Forest
0.55
74.7396
25.2604
10-fold
cross-
validation
9
768
0.4313
0.3105
0.653
0.791
Trees
Naïve
Bayes
0.03
76.3021
23.6979
10-fold
cross-
validation
9
768
0.4664
0.2841
0.678
0.802
Bayes
SVM
0.09
77.3438
22.6563
10-fold
cross-
validation
9
768
0.4682
0.2266
0.740
0.785
Functions
Neural
Networks
0.81
75.1302
24.8698
10-fold
cross-
validation
9
768
0.4445
0.2938
0.653
0.799
Functions
JRip
0.19
74.4792
25.5208
10-fold
cross-
validation
9
768
0.4171
0.3461
0.659
0.780
Rules
Decision
Tree (J48)
0.14
73.8281
26.1719
10-fold
cross-
validation
9
768
0.4164
0.3158
0.632
0.790
Tree
Time is the TIME taking to build the model
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 134
MAE (Mean Absolute Error) is a measure of how
close forecast or predictions are to the eventual
outcome.
Kappa Statistic is a metric that compares an
observed accuracy with an expected accuracy
(Random Chance)
YES means tested positive to diabetes. NO means
tested negative for diabetes
Table 4 shows the results with 6 attributes of the
classification and comparison of the various
machine leaning algorithms and parameters
considered.
Table 4: Comparison of various classification algorithms with smaller data set and less attributes
Algorithm
Time
Correctly
Classified
%
Incorrectly
Classified
%
Test
Mode
Attributes
No of
instances
Kappa
statistic
MAE
Precision
of YES
Precision
of NO
Classific-
ation
Decision
Table
0.09
67.9688
32.0313
10-fold
cross-
validation
6
384
0.3748
0.3101
0.581
0.734
Rules
Random
Forest
0.42
71.875
28.125
10-fold
cross-
validation
6
384
0.3917
0.3438
0.639
0.761
Trees
Naïve Bayes
0.01
70.5729
29.4271
10-fold
cross-
validation
6
364
0.352
0.3297
0.633
0.739
Bayes
SVM
0.04
72.9167
27.0833
10-fold
cross-
validation
6
384
0.3837
0.2708
0.711
0.735
Functions
Neural
Networks
(Perceptron)
0.17
59
41
10-fold
cross-
validation
6
384
0.1156
0.4035
0.444
0.672
Functions
JRip
0.01
64
36
10-fold
cross-
validation
6
384
0.2278
0.4179
0.514
0.714
Rules
Decision
Tree (J48)
0.03
64 %
36
10-fold
cross-
validation
6
384
0.1822
0.4165
0.519
0.685
Tree
Time is the TIME taking to build the model.
MAE (Mean Absolute Error) is a measure of how
close forecast or predictions are to the eventual
outcome.
Kappa Statistic is a metric that compares an
observed accuracy with an expected accuracy
(Random Chance)
YES means tested positive to diabetes. NO means
tested negative for diabetes
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 135
Table 5 and 6: Ranking of Precision of Positive Diabetes and Negative Diabetes using different algorithms
showing smaller and larger data sets respectively
Smaller Dataset 384
Large Data Set 768
Algorithm
Precision of
YES (Positive
Diabetes)
Precision of
NO (Negative
Diabetes)
Algorithm
Precision of
YES (Positive
Diabetes)
Precision of
NO (Negative
Diabetes
SVM
0.711
0.735
SVM
0.74
0.785
Random Forest
0.639
0.761
Naïve Bayes
0.678
0.802
Naïve Bayes
0.633
0.739
JRip
0.659
0.78
Decision Table
0.581
0.734
Random Forest
0.653
0.791
Decision Tree
(J48)
0.519
0.685
Neural Networks
(Perceptron)
0.653
0.799
JRip
0.514
0.714
Decision Tree (J48)
0.632
0.79
Neural Networks
(Perceptron)
0.444
0.672
Decision Table
0.619
0.771
Table 7 and 8: Ranking of Correctly Classified and Incorrectly Classified with the time to build the model
showing smaller and larger data sets respectively using different algorithm.
Smaller Dataset 384
Large Data Set 768
Algorithm
Time
Correctly
Classified
Incorrectly
Classified
Algorithm
Time
Correctly
Classified
Incorrectly
Classified
SVM
0.04
sec
72.92%
27.08%
SVM
0.09
sec
77.34%
22.66%
Random Forest
0.42
sec
71.88%
28.13%
Naïve Bayes
0.03
sec
76.30%
23.70%
Naïve Bayes
0.01
sec
70.57%
29.43%
Neural
Networks
(Perceptron)
0.81
sec
75.13%
24.87%
Decision Table
0.09
sec
67.97%
32.03%
Random Forest
0.55
sec
74.74%
25.26%
JRip
0.01
sec
64%
36%
JRip
0.19
sec
74.48%
25.52%
Decision Tree
(J48)
0.03
sec
64%
36%
Decision Tree
(J48)
0.14
sec
73.83%
26.17%
Neural
Networks
(Perceptron)
0.17
sec
59%
41%
Decision Table
0.23
sec
72.40%
27.60%
Table 9Descriptive Analysis of various Datasetattributes
Attribute number
Mean
Standard Deviation
1
3.8
3.4
2
120.9
32.0
3
69.1
19.4
4
20.5
16.0
5
79.8
115.2
6
32.0
7.9
7
0.5
0.3
8
33.2
11.8
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 136
B. Discussion
Table3 shows the comparison of the result for 768
instances and 9 attributes. It was observed that all
the algorithms have higher Kappa statistic compared
to MAE (Mean Absolute Error). Also, correctly
classified instances are higher than incorrectly
classified instances. This is an indication that with
higher data sets, the predictive analysis is more
reliable. SVM and NBrequire large sample size in
order to achieve maximum prediction accuracy as
shown in the table 3, while Decision Tree and
Decision Table have the least precision.
Table4 shows the comparison of the result for 384
instances and 6 attributes. The Kappa statistics for
Neural Networks, JRip and J48 are lower compared
to MAE and this does not portray precision and
accuracy. This shows that with smaller datasets
Neural Networks, JRip and J48 shows drastic
reduction in the percentage of correctly classified
instances in comparison to incorrectly classified
instances. However, with smaller data set SVM and
RF shows high accuracy and precision. Whereas
Decision Table built the model with more time
compare to JRip and Decision Tree. Therefore, less
time does not guarantee accuracy. If Kappa Statistic
is less than Mean Absolute Error (MAE), the
algorithm will not show precision and accuracy. It
follows that, the algorithm which such
characteristics cannot be used for that data set as it
will not show precision and accuracy.
Table 6shows precision for larger data set and
smaller data set with SVM reflecting the algorithm
with highest prediction. Also table 5 shows SVM
being the algorithm with highest precision. Smaller
data sets.
Tales 7 and 8 shows the comparison of percentage
of correctly classified and incorrectly classified for
smaller and large datasets respectively with the time
to build the model. From Table 7,the results reveal
Naive Bayes and JRip as the algorithms with fastest
time to build, however the percentage of correctly
classified is lower in JRip which shows that Time to
build as model is not tantamount to accuracy. In the
same vein, SVM has the highest level of accuracy
with time of 0.04 seconds. Comparing this results
with Table 8Neural Networks (Perceptron) was the
third correctly classified algorithm. This means that
Neural Network performs well with large dataset as
compared to small data set. Also, the results shows
that Decision Table does not perform well with
large dataset. By and large, SVM algorithm shows
the highest classification and the larger the dataset,
the higher the precision.
Table 9 shows the mean and standard deviation of
all the attributes used in this research reveals that
Plasma glucose concentration (attribute 2) has the
highest mean as well as Diabetes pedigree function
(attribute 7) with the lowest mean which is an
indication of strong influence on small data set.
However, a lower Standard Deviation (SD) is not
necessarily more desirable which means Diabetes
pedigree function (attribute 7) might not be of
significance vale when analyzing large data set.
V. CONCLUSION AND
RECOMMENDATION FOR FURTHER
WORKS
ML classification requires thorough fine tuning
of the parameters and at the same time sizeable
number of instances for the data set. It is not a
matter of time to build the model for the algorithm
only but precision and correct classification.
Therefore, the best learning algorithm for a
particular data set, does not guarantee the precision
and accuracy for another set of data whose attributes
are logically different from the other. However, the
key question when dealing with ML classification is
not whether a learning algorithm is superior to
others, but under which conditions a particular
method can significantly outperform others on a
given application problem. Meta-learning is moving
in this direction, trying to find functions that map
datasets to algorithm performance [12]. To this end,
meta-learning uses a set of attributes, called meta-
attributes, to represent the characteristics of learning
tasks, and searches for the correlations between
these attributes and the performance of learning
algorithms. Some characteristics of learning tasks
are: the number of instances, the proportion of
categorical attributes, the proportion of missing
values, the entropy of classes, etc.
[3]provided an extensive list of information
and statistical measures for a dataset. After a better
understanding of the strengths and limitations of
each method, the possibility of integrating two or
more algorithms together to solve a problem should
be investigated. The objective is to utilize the
strengths of one method to complement the
weaknesses of another. If we are only interested in
the best possible classification accuracy, it might be
difficult or impossible to find a single classifier that
performs as well as a good ensemble of
classifiers.SVM, NB and RF machine learning
algorithms can deliver high precision and accuracy
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 137
regardless of the number of attributes and data
instances. This research shows that time to build a
model is one factor on one hand; and precision with
kappa statistic while MAE is another factor on the
other hand. Therefore, ML algorithms requires
precision, accuracy and minimum error to have
supervised predictive machine learning.
This work recommends that for large data sets,
a distributed processing environment should be
considered. This will create room for high level of
correlation among the variables which will
ultimately make the output of the model more
efficient.
REFERENCES
[1] Alex S.& Vishwanathan, S.V.N. (2008). Introduction to
Machine Learning. Published by the press syndicate of the
University of Cambridge, Cambridge, United Kingdom.
Copyright ⓒ Cambridge University Press 2008. ISBN: 0-
521-82583-0. Available at KTH website:
https://www.kth.se/social/upload/53a14887f276540ebc81ae
c3/online.pdf Retrieved from website:
http://alex.smola.org/drafts/thebook.pdf
[2] Bishop, C. M. (1995). Neural Networks for Pattern
Recognition. Clarendon Press, Oxford, England. 1995.
Oxford University Press, Inc. New York, NY, USA ©1995
ISBN:0198538642 Available at:
http://cs.du.edu/~mitchell/mario_books/Neural_Networks_f
or_Pattern_Recognition_-_Christopher_Bishop.pdf
[3] Brazdil P., Soares C. &da Costa, J. (2003).
Ranking Learning Algorithms: Using IBL
and Meta-Learning on Accuracy and Time
Results.Machine LearningVolume 50,
Issue 3,2003.Copyright ©Kluwer
Academic Publishers. Manufactured in The Netherlands,
doi:10.1023/A:1021713901879pp. 251–277. Available at
Springer website:
https://link.springer.com/content/pdf/10.1023%2FA%3A10
21713901879.pdf
[4] Cheng, J., Greiner, R., Kelly, J., Bell, D.& Liu, W. (2002).
Learning Bayesian networks from data: An information-
theory based approach. Artificial Intelligence Volume 137.
Copyright © 2002. Published by Elsevier Science B.V. All
rights reserved pp. 43 – 90. Available at science Direct:
http://www.sciencedirect.com/science/article/pii/S00043702
02001911
[5] Domingos, P. & Pazzani, M. (1997). On the optimality of
the simple Bayesian classifier under zero-one loss. Machine
Learning Volume 29, pp. 103–130 Copyright © 1997
Kluwer Academic Publishers. Manufactured in The
Netherlands. Available at University of Trento website:
http://disi.unitn.it/~p2p/RelatedWork/Matching/domingos9
7optimality.pdf
[6] Elder, J. (n.d). Introduction to Machine Learning and
Pattern Recognition. Available at LASSONDE University
EECS Department York website:
http://www.eecs.yorku.ca/course_archive/2011-12/F/4404-
5327/lectures/01%20Introduction.pd
[7] Good, I.J. (1951). Probability and the Weighing of
Evidence, Philosophy Volume 26, Issue 97, 1951. Published
by Charles Griffin and Company,
London 1950.Copyright © The Royal Institute of
Philosophy 1951,pp. 163-164.doi:
https://doi.org/10.1017/S0031819100026863. Availableat
Royal Institute of Philosophy website:
https://www.cambridge.org/core/journals/philosophy/article
/probability-and-the-weighing-of-evidence-by-goodi-j-
london-charles-griffin-and-company-1950-pp-viii-119-
price-16s/7D911224F3713FDCFD1451BBB2982442
[8] Hormozi, H., Hormozi, E. & Nohooji, H. R. (2012). The
Classification of the Applicable Machine Learning Methods
in Robot Manipulators. International Journal of Machine
Learning and Computing (IJMLC), Vol. 2, No. 5, 2012
doi: 10.7763/IJMLC.2012.V2.189pp. 560 – 563.
Available at IJMLC website:
http://www.ijmlc.org/papers/189-C00244-001.pdf
[9] Kotsiantis, S. B. (2007). Supervised Machine Learning: A
Review of Classification Techniques. Informatica 31
(2007). Pp. 249 – 268. Retrieved from IJS website:
http://wen.ijs.si/ojs-
2.4.3/index.php/informatica/article/download/148/140.
[10] Lemnaru C. (2012). Strategies for dealing with Real World
Classification Problems, (Unpublished PhD thesis) Faculty
of Computer Science
and Automation, Universitatea Technica,
Din Cluj-Napoca. Available at website:
http://users.utcluj.ro/~cameliav/documents/TezaFinalLemna
ru.pdf
[11] Logistic Regression pp. 223 – 237. Available at:
https://www.stat.cmu.edu/~cshalizi/uADA/12/lectures/ch12
.pdf
[12] Neocleous C. & Schizas C. (2002). Artificial Neural
Network Learning: A Comparative Review. In: Vlahavas
I.P., Spyropoulos C.D. (eds)Methods and Applications of
Artificial Intelligence. Hellenic Conference on Artificial
IntelligenceSETN 2002. Lecture Notes in Computer
Science, Volume 2308. Springer, Berlin, Heidelberg, doi:
10.1007/3-540-46014-4_27 pp. 300-313. Available at:
https://link.springer.com/chapter/10.1007/3-540-46014-
4_27
.
[13] Newsom, I. (2015). Data Analysis II:
Logistic Regression. Available at:
http://web.pdx.edu/~newsomj/da2/ho_logistic.pdf
[14] Nilsson, N.J. (1965). Learning machines. New York:
McGraw-Hill.Published in: Journal of IEEE Transactions
on Information Theory Volume 12 Issue 3, 1966. doi:
10.1109/TIT.1966.1053912 pp. 407 – 407. Available at
ACM digital library website:
http://dl.acm.org/citation.cfm?id=2267404
[15] Pradeep, K. R. & Naveen, N. C. (2017). A Collective Study
of Machine Learning (ML)Algorithms with Big Data
Analytics (BDA) for Healthcare Analytics (HcA).
International Journal of Computer Trends and Technology
(IJCTT) – Volume 47 Number 3, 2017. ISSN: 2231-2803,
doi: 10.14445/22312803/IJCTT-V47P121, pp 149 – 155.
Available from IJCTT website:
http://www.ijcttjournal.org/2017/Volume47/number-
3/IJCTT-V47P121.pdf
[16] Rob Schapire (n.d) Machine Learning Algorithms for
Classifrication.
[17] Rosenblatt, F. (1962), Principles of Neurodynamics.
Spartan, New York.
[18] Setiono R. and Loew, W. K. (2000), FERNN: An algorithm
for fast extraction of rules from neural networks, Applied
Intelligence.
International Journal of Computer Trends and Technology (IJCTT) – Volume 48 Number 3 June 2017
ISSN: 2231-2803 http://www.ijcttjournal.org Page 138
[19] Shai Shalev-Shwartz and Shai Ben-David (2014).
Understanding Machine Learning From Theory to
Algorithms.
[20] T. Hastie, R . Tibshirani, J. H. Friedman (2001) ― The
elements of statistical learning,‖ Data mining, inference,
and prediction, 2001, New York: Springer Verlag.
[21] Taiwo, O. A. (2010). Types of Machine Learning
Algorithms, New Advances in Machine Learning, Yagang
Zhang (Ed.), ISBN: 978-953-307-034-6, InTech, University
of Portsmouth United Kingdom. Pp 3 – 31. Available at
InTech open website:
http://www.intechopen.com/books/new-advances-in-
machine-learning/types-of-machine-learning-algorithms
[22] Tapas Kanungo, D. M. (2002). A local search
approximation algorithm for k-means clustering.
Proceedings of the eighteenth annual symposium on
Computational geometry. Barcelona, Spain: ACM Press.
[23] Timothy Jason Shepard, P. J. (1998). Decision Fusion
Using a Multi-Linear Classifier. In Proceedings of the
International Conference on Multisource-Multisensor
Information Fusion.
[24] Vapnik, V. N. (1995). The Nature of Statistical Learning
Theory. (2nd ed.).
Springer Verlag. Pp. 1 – 20.
Retrieved from website:
https://www.andrew.cmu.edu/user/kk3n/simplicity/vapnik2
000.pdf
[25] Witten, I. H. & Frank, E. (2005). Data Mining: Practical
machine learning tools and techniques (2nd ed.), ISBN: 0-
12-088407-0, Morgan Kaufmann Publishers, San Francisco,
CA, U.S.A. © 2005 Elsevier Inc.Retrieved from website:
ftp://93.63.40.27/pub/manuela.sbarra/Data Mining Practical
Machine Learning Tools and Techniques - WEKA.pdf
