Abstract:
In Ethiopia, diabetes disease is a major health problem due to life style, lack of health care physician and laboratory facilities. Nowadays, knowledge-based system became a choice for supporting physicians in decision making regardless of diagnosis and recommended for various diseases. Both rule-based and case-based reasoning are the main method used to develop the knowledge-based reasoning. The main objective of this study is to develop an intelligent knowledge-based system for diagnosis and treatment of diabetes employing both rule-based and case-based reasoning. The two major tasks of data mining, predictive and descriptive modeling are used for building a knowledge-based system with rule-based reasoning and case-based reasoning respectively. While rule-based reasoning system is developed using Prolog programming language, case-based system is developed by JCOLLIBIR studio with its CBR main cycle (Retrieval, Reuse, Revise and retain). Design science research methodology is followed in this study. Hidden knowledge is mined from the diabetes dataset which is collected from Haramaya University Hiwot Fana Specialized hospital, Haramaya general hospital and Harari general hospital. The collected dataset was preprocessed for removing duplicates, missing values, normalized data, outliers, noisy and errors. The predictive model is experimented using J48 decision tree, PART and JRip rule induction algorithms, whereas the descriptive model is experimented using K-means and Farthest first. Finally, the result of the JRip algorithm is used to construct the rule base, whereas K-means result is used for designing the case base automatically since they achieved a better experimental performance. A rule dominant approach is used to integrate rule based with case based reasoning to develop an intelligent knowledge based system that means rule based reasoning followed case based reasoning which has decision module in between them. The intelligent knowledge based system registered 95.23% accuracy with an average Precision and Recall of 100% and 89.7% respectively. Similarly, the user acceptance testing registered 93.2%. This shows the system has registered a promising result. However, RBR reasoning depends on domain expert to define rule accurately and has no self-learning feature which needs further improvement in the future.
