International Journal of Research and Innovation in Applied Science (IJRIAS) | Volume VII, Issue VIII, July 2022 | ISSN 2454–6194
Heart Disease Predictive Model Using Filter-Based Selection Techniques and Tree-Like Classifiers
Awe Oluwayomi1, Aiyeniko Olukayode2, Adedokun Olufemi Adewale3, Funso Bukola Omolara4,
and Samuel Ruth Medinat5
1Department of Computer Science, University of Lagos, Lagos State, Nigeria
2Department of Computer Science, Lagos State University, Lagos State, Nigeria
3Department of Computer Science, University of Ilorin, Ilorin, Nigeria
4,5Department of Computer Science Kogi State Polytechnic, Kogi State, Nigeria
Abstract: The attribute selection is considered a major phase that eliminates redundant attributes thereby improving the accuracy of the predictive or diagnostic model. Designing a model with unrelated attributes may influence the accuracy or result in more memory space used during diagnosis or prediction. This paper examined the impact of the filter-based attribute selection technique on the heart disease diagnostic model. Three filter-based techniques: Relief-F, Information Gain and Chi-square were applied to the heart disease dataset. Five tree-like learning algorithms: ID3 (Iterative Dichotomiser 3), C4.5 Decision Tree, Reptree (RP), Random Forest (RF), Classification and Regression Tree (CART) were applied to classify the reduced attributes. The experimental results in terms of accuracy, precision and recall showed that the relief-f attribute selection outperformed information gain and chi-square with the best predictive accuracy of 93.4983% in IDE, the precision value of 0.93500 in IDE and recall value of 0.93500 in IDE classifier.
Keywords: Chi-square, Data Mining, Filter-based, Relief-F, Information Gain
I.INTRODUCTION
Several deadly ailments affect humans, one among these diseases is heart disease [1]. Heart disease is common sickness in adults, this has currently increased the death rate over the globe [2]. Heart disease is an illness that affects the circulatory system of the heart [3]. The medical domain is surrounded by huge data but inadequate action has been given to this data in proffering answers to some life-threatening problems such as diagnosis of diseases. Available are many approaches to accomplish this task, but data mining remains the most significant method [4]. The approach of data mining uses data to methodically or logically discover inadequacies, ease costs and improve upon best practices in medicine [5]. Computational intelligent systems or data mining are tools that can be employed to perform predictions with medical datasets with many responses. Heart disease prediction is a difficult process that necessitates being achieved correctly and creditably [6]. Conclusions made by physicians may occasionally be centered on instinct rather than on the unknown information in a patient’s data, this sometimes results in unwanted flaws and expensive costs in healthcare which also constitute adverse effects on the standard of service given to patients.