Content uploaded by Ismail Olaniyi Muraina
Author content
All content in this area was uploaded by Ismail Olaniyi Muraina on May 07, 2022
Content may be subject to copyright.
209
209
5th INTERNATIONAL AFRICAN CONFERENCE ON CURRENT STUDIES
February 2-5, 2022 Cairo, Egypt
MACHINE LEARNING ALGORITHMS SELECTION PROBLEMS RESOLVED
WITH ARTIFICIAL INTELLIGENT RULE-BASED EXPERT SYSTEM USING
VISIRULE
Ismail Olaniyi MURAINA
Computer Science Department, School of Science
Adeniran Ogunsanya College of Education, Lagos Nigeria.
ORCID ID: 0000-0002-9633-6080
Moses Adeolu AGOI
Computer Science Department, School of Science
Adeniran Ogunsanya College of Education, Lagos Nigeria.
ORCID ID: 0000-0002-8910-2876
Abstract
Everybody is confronted daily with cluster of decisions that must be appropriately taken in
the process of making decision; individuals are faced with and most often fall prey to series
of common biases, fallacies, and many other decision-making odds. In determining which
algorithm to apply for analysis (with machine learning using supervised and unsupervised
approaches) open to critical steps to be taken and also highly depend on many factors ranging
from the type of problem at hand to the expected outcomes. The study looks at how artificial
intelligent approach with expert system would be helpful in making timely decision on which
type of algorithms is capable to be applied and implemented to have desired results. The
study also uses VisiRule software to model series of successful channels to arrive at a good
decision-making means. The use of VisiRule (Artificial Intelligent Based Expert System) was
employed to give directional path ways to the selection of appropriate algorithms from
supervised and unsupervised machine learning to different classification methods, regression
methods, clustering approaches, dimensionality reduction methods, and association rules. The
outcome of this study demonstrates the easy way through paths to select relevant and most
appropriate model or algorithm that best fit the analysis at hand with detailed explanation of
each alternative option. The use of VisiRule software has proven the easy way to achieve
decision making problems without any codes requirement for such actions. Decision making
challenges could be resolved by just implementing artificial intelligent rule-based expert
system which require less time, coding free, and highly achievable accurate outcomes.
Keywords: Decision Making, VisiRule, Artificial Intelligent, Rule-based System, Expert
System
Introduction
Decision making has been a carefully and procedural actions taken by every individual in the
society. Researchers are increasingly investigating the conditions attached to application of
models or algorithms and how to identify the most suitable existing algorithms/models for
solving a problem. The decision-making regarding selection of Algorithm/models is
concerned with the kind of procedures or conditions available to apply an algorithm and
selecting the best algorithm to solve a given problem (Kotthoff, 2012). Researchers of
machine learning algorithms need methods that can help them to identify algorithm or their
210
210
5th INTERNATIONAL AFRICAN CONFERENCE ON CURRENT STUDIES
February 2-5, 2022 Cairo, Egypt
groupings (combinations) that achieve the potentially best performance. Selecting the best
algorithm to solve a given problem has to do with having well conversant knowledge of
conditions to select or use an algorithm/model and which one (Out of available algorithms)
could give the required and optimal solution to the problem at hand and finally at what mode
of operation to follow – Single or combined method (Abdulrahman, Adamu, Ibrahim, &
Muhammad, 2017).
A proliferation of algorithms/models exist, rooted in the fields of machine learning, statistics,
pattern recognition, artificial intelligence, and database systems, which are used to perform
different data analysis jobs on large volumes of data. The decision to take in order to
recommend the most suitable algorithms has thus become rather challenging. Moreover, the
problem is exacerbated by the fact that it is necessary to consider different combinations of
parameter settings, or the constituents of composite methods such as ensembles
(Abdulrahman, Adamu, Ibrahim, & Muhammad, 2017).
It was observed that before a machine learning algorithm/model is trained, the researcher of a
machine learning software tool or algorithm typically must manually select a machine
learning algorithm and set one or more model parameters termed hyper-parameters. The
algorithm and hyper - parameter values used can greatly impact the resulting model’s
performance, but their selection requires special expertise as well as many labor-intensive
manual iterations. To make machine learning accessible to everyone interested to use them,
with limited computing expertise, computer science researchers have proposed various
automatic selection methods for algorithms and/or hyperparameter values for a given
supervised machine learning problem (Luo, Gang, 2017).
The correct use of model evaluation, model selection, and algorithm selection techniques is
vital in academic machine learning research as well as in many industrial settings (Dhabarde,
2019). Selecting the right algorithm is an important problem in computer science, because the
algorithm often has to exploit the structure of the input to be efficient. So, solutions to the
algorithm selection problem can inspire models of human strategy selection. Therefore, the
algorithm selection problem as a special case of meta-reasoning and need to be tackled in a
systematically approach manner (Lieder, Plunkett, Hamrick, Russell, Hay, & Griffiths, 2014).
Related Literatures
The approach to select appropriate algorithm/model was viewed in two ways: the first aspect
is looking at conditions to use an algorithm/model or combinations of algorithms/models and
how possible for researcher to select the best algorithm/model to solve a kind of problem.
Studies in the past contributed to this trend. The use of application software is a great
opportunity to resolve the problem with artificial intelligent rule-based expert system via
Visirule. This knowledge can help us to select the best algorithm for these instances.
According to Kotthoff, (2012) Algorithm Selection techniques have achieved significant
performance improvements. They unified and organized the vast literature according to
criteria that determine Algorithm Selection systems in practice. The comprehensive
classification of approaches identified and analyzed the different directions from which
Algorithm Selection has been approached. Their paper contrasted and compared different
methods for solving the problem as well as ways of using these solutions. Their study was
closed by identifying directions of current and future research.
This survey presented by Abdulrahman, Salisu Mamman; Adamu, Alhassan; Ibrahim, Yazid
Ado & Muhammad, Akilu Rilwan (2017) looked into an overview of the contributions made
211
211
5th INTERNATIONAL AFRICAN CONFERENCE ON CURRENT STUDIES
February 2-5, 2022 Cairo, Egypt
in the area of algorithm selection problems. They presented different methods for solving the
algorithm selection problem identifying some of the future research challenges in this
domain. They further added that researchers have long ago recognized that it is difficult to
identify a single best algorithm that will give the best performance across all problems. This
is why later on many researchers have developed different approaches to addressing the
algorithm selection problems. There are many approaches to addressing the algorithm
selection problem; in connection to this, Kotthoff, Gent & Miguel, (2012) claimed that
machine learning is an established method of selecting algorithms to solve hard search
problems.
The algorithm selection problem, as explained by Rice, (1976) has attracted a great deal of
attention, as it endeavours to select and apply the best algorithm(s) for a given task (Brazdil,
Carrier, Soares & Vilalta, 2008; Smith-Miles, 2009). The algorithm selection problem can be
cast as a learning problem: the aim is to learn a model that captures the relationship between
the properties of the datasets, or meta-data, and the algorithms, in particular their
performance. This model can then be used to predict the most suitable algorithm for a given
new dataset as viewed by Abdulrahman, Salisu Mamman; Adamu, Alhassan; Ibrahim, Yazid
Ado & Muhammad, Akilu Rilwan (2017).
Kotthoff, Lars; Gent, Ian P & Miguel, Ian (2012) conducted a study where they compared the
performance of a large number of different machine learning techniques from different
machine learning methodologies on five data sets of hard algorithm selection problems from
the literature. They demonstrated that there is significant scope for improvement both
compared with existing systems and in general. At the end, they gave clear recommendations
as to which machine learning techniques were likely to achieve good performance in the
context of algorithm selection problems. In particular, they showed that linear regression and
alternating decision trees have a very high probability of achieving better performance than
always selecting the single best algorithm. Luo, (2017) researched on machine learning
studies automatic algorithms that improve themselves through experience. Their paper
reviewed methods, identified several of their limitations in the big biomedical data
environment, and provided preliminary thoughts on how to address these limitations. The
findings established a foundation for future research on automatically selecting algorithms
and hyper-parameter values for analyzing big biomedical data.
Guo & Hsu, (2007) In their paper, they presented a machine learning-based approach to
address models induced from algorithmic performance data can represent the knowledge of
how algorithmic performance depends on some easy-to-compute problem instance
characteristics. Using these models, they could estimate approximately whether an input
instance was exactly solvable or not. Furthermore, when it was classified as exactly
unsolvable, they could select the best approximate algorithm for it among a list of candidates.
The results showed that the machine learning-based algorithm selection system could
integrate both exact and inexact algorithms and provide the best overall performance
comparing to any single candidate algorithm
Dhabarde, (2019) regarded machine learning as subfield of AI concerned with intelligent
systems that learn. According to him, to understand machine learning, it is helpful to have a
clear notion of intelligent systems. Therefore, their paper reviewed different techniques that
could be used for each of the three subtasks and discussed the main advantages and
disadvantages of each technique with references to theoretical and empirical studies.
Common cross-validation techniques such as leave-one- out cross-validation and k-fold
cross-validation were reviewed, the bias-variance trade-off for choosing k was discussed, and
practical tips for the optimal choice of k were given based on empirical evidence
212
212
5th INTERNATIONAL AFRICAN CONFERENCE ON CURRENT STUDIES
February 2-5, 2022 Cairo, Egypt
Lieder et al., (2014) applied theory to model how people choose between cognitive strategies
and test its prediction in a behavioral experiment. They found out that people quickly learn to
adaptively choose between cognitive strategies. People’s choices in our experiment were
consistent with the model used but inconsistent with previous theories of human strategy
selection. Rational meta-reasoning appears to be a promising framework for reverse-
engineering how people select between cognitive strategies and translating the results into
better solutions to the algorithm selection problem. Masood, Khan, Hussain & Shaukat,
(2020) carried out studies on the systematic literature review (SLR) that has been performed
to get 20 studies (2012-2019) in the area of EDM. From these studies, 11 highly advanced
machine learning models has been obtained and they have implemented them on 2 public
student databases in order to predict their future outcomes. Feature extraction techniques
were applied and then models have been trained based on the databases to get the required
results. Results of different machine learning models were compared in order to find out the
best model among them based on accuracy and F-measure. With these experiments, weak
students can be easily identified and proper precautions can be taken in order to help them.
Materials and Methods
The study employed the use of artificial intelligent rule-based expert system using Visirule
software. Visirule software is designed for researchers as a decision supporting tool that the
rules are basically and precisely presented without writing any simple code. The approach
was based on the use of Logic Programming Model. The rule-based Expert system is of great
use to researchers in making appropriate and relevant selection of machine learning algorithm
suitable for statistical data analysis in researches (Muraina, Rahman, Adeleke, & Aiyegbusi,
2013). It allows researchers to concentrate on explaining and establishing the structure of the
logic correctly using their chosen tools - those embedded materials that can assist researcher
to accomplish his mission (Spenser, 2007; Bilgi, kulkarni, & Spenser, 2010).
213
213
5th INTERNATIONAL AFRICAN CONFERENCE ON CURRENT STUDIES
February 2-5, 2022 Cairo, Egypt
Analysis and Results
Figure 1: Visirule showing easy selection of either supervised or unsupervised machine
learning
The figure 1 displays the condition to choose either supervised or unsupervised machine
learning algorithm likewise it goes further to ask another question until the right decision is
made on the suggestions provided by the software
214
214
5th INTERNATIONAL AFRICAN CONFERENCE ON CURRENT STUDIES
February 2-5, 2022 Cairo, Egypt
Figure 2: Visirule showing the list of regression types
The second figure (Figure 2) depicts the collection of regression analyses, from which the
researcher would choose one. From the figure 2, the details of each of the types are shown at
right hand side with applicable conditions for their selection
Discussion
The use of the artificial intelligent rule-based expert system showed the easy way to
determine which of the algorithm to use – based on the condition of its use. The figure 1 and
2 showed the systematic procedure to best select the algorithms/model for further analysis
and decision making. This approach is said to be useful and cost effective in decision making
rather than manual selection method. The software as well will generate codes that can be
export to the web and other format. The application of the Visirule covered both supervised
and unsupervised machine learning algorithms/models
Conclusion
For years, selecting the best algorithm/model to solve a given problem has been the subject of
many studies. In this paper, we have covered briefly the use of artificial intelligent rule-based
expert system to select appropriate algorithm by first considering the conditions attached to
the use and to select the best algorithm or group of algorithms that can perform better among
others. Also, in this research paper, we have performed systematic literature review of
machine learning models.
215
215
5th INTERNATIONAL AFRICAN CONFERENCE ON CURRENT STUDIES
February 2-5, 2022 Cairo, Egypt
References
Abdulrahman, Salisu Mamman; Adamu, Alhassan; Ibrahim, Yazid Ado & Muhammad, Akilu
Rilwan (2017) An Overview of the Algorithm Selection Problem. International Journal
of Computer (IJC) (26)1, 71-98
Brazdil, P., Carrier, C. G., Soares, C., & Vilalta, R. (2008). Metalearning: Applications to
data mining. Springer Science & Business Media.
Dhabarde, Swati (2019) Approach towards Model Evaluation, Model Selection, and
Algorithm Selection in Machine Learning. Pramana Research Journal (9)6, 2019, 396 -
408
Guo, Haipeng & Hsu, William H (2007) A machine learning approach to algorithm selection
for NP-hard optimization problems: a case study on the MPE problem. Ann Oper Res
(August, 2007) 156: 61–82
Kotthoff, Lars (2012) Algorithm Selection for Combinatorial Search Problems: A survey
Kotthoff, Lars; Gent, Ian P & Miguel, Ian (2012) An Evaluation of Machine Learning in
Algorithm Selection for Search Problems
Lieder, Falk; Plunkett, Dillon; Hamrick, Jessica B; Russell, Stuart J; Hay, Nicholas J &
Griffiths, Thomas L. (2014) Algorithm selection by rational meta-reasoning as a model
of human strategy selection
Luo, Gang (2017) A Review of Automatic Selection Methods for Machine Learning
Algorithms and Hyper parameter Values
Masood, Muhammad Faisal; Khan, Aimal; Hussain, Farhan & Shaukat, Arslan (2020)
Towards the Selection of Best Machine Learning Model for Student Performance
Analysis and Prediction
Rice, J. R. (1976). The algorithm selection problem. Advances in computers, 15, 65-118.
Smith-Miles, K. A. (2009). Cross-disciplinary perspectives on meta-learning for algorithm
selection. ACM Computing Surveys (CSUR), 41(1), 6.
