Developing Stroke Diagnosis and Treatment Knowledge-based System By Integrating Expert Knowledge and Machine Learning

Abstract

Stroke is a medical condition where the blood arteries in the brain rupture, causing brain damage. This interruption of blood flow can lead to the development of symptoms. According to the World Health Organization (WHO), stroke is the leading cause of death and disability worldwide. Recognizing the warning signs of a stroke early on can help lessen the severity of the condition. To the best of the author’s knowledge, no research has been conducted on creating a knowledge-based system that combines expert knowledge and machine learning prediction to assist healthcare workers in managing stroke without the need for experts. This study aims to develop such a system, referred to as a knowledge- based system (KBS), for stroke diagnosis and treatment by integrating machine learning prediction with expert knowledge. Data was gathered from Debre Birhan Referral Hospital (DBRH) using in-depth interviews with experts selected through purposive sampling techniques and from a public dataset obtained from the Keggle website. The dataset comprised 5,110 instances, 12 attributes, and 2 class labels. To balance the class labels, a Synthetic Minority Over-sampling Technique (SMOTE) was utilized, increasing the number of instances from 5,110 to 9,720 for experimentation. This research utilized the Design Science Research Methodology. Expert knowl- edge was extracted and represented using production rules, which were then mod- eled using a decision tree. To identify the most suitable machine learning classifier models, 9 experiments were conducted with a decision tree, random forest, and support vector machine classifiers, employing 10-fold cross-validation and the per- centage split test option in two scenarios: one using all attributes and another using selected attributes. Finally, the rules of the random forest classifier with the selected attributes achieved the best performance, with an accuracy of 99%, and were integrated with expert knowledge to develop the knowledge-based system. The collaboration of two programming languages, Python and Prolog, was em- ployed to develop the knowledge-based system. The rule base was constructed using the Prolog programming language and SWI-Prolog 7.6.4, while HTML and xiii Abbreviations xiv a sublime text editor were used to create the system’s graphical user interface (GUI). The developed knowledge-based system was evaluated by preparing test cases and user acceptance testing, achieving a performance rating of 95% and a user acceptance score of 88%. Therefore, the knowledge-based system successfully fulfills its intended purpose without needing experts.