An Integration Of Case Based Reasoning With Data Mining Technique To Enhance The Effectiveness

Abstract

Nowadays nothing is more worth than health. It is the most important aspects of human life. Pneumonia is one of the health related problem which is the most killing disease in Ethiopia even in Africa. The reason for its high level of killing is the difficulty of the disease nature to treat as well as the shortage of health professionals. We examined the strengths and weaknesses of various reasoning paradigms including case-based reasoning, rule-based reasoning and data mining techniques. We discuss how to combine them to form a more robust and better-performing hybrid. In a decision support system to address the variety of tasks a user performs, a single type of knowledge and reasoning method is often not sufficient. A combination of different methods has often shown the best results. In this study Case Based Reasoning was mixed with data mining techniques and Rule Based Reasoning approaches to promote synergies and benefits beyond those achievable using CBR or other individual reasoning approaches alone. The dataset is collected from Adama medical health science collage and hospital and different standard guideline using document review technique. The data from Medical Record Numbers are collected and analyzed to be used in Case Based Reasoning. The dataset has 1007 records and 15 fields. The industry standard CRISP-DM data mining process model used throughout this research for the purpose of preprocessed and model building. The case stored in the database, preprocessed using a data mining techniques. The dataset is further used to generate classification rule and association rule. To develop a prototype, we use mainly jcolibri frame work library for CBR prototype, Eclipse IDE, JavaFx, FXML and Navicat (MySQL database). The evaluation of the prototype is done from two system measurement perspectives. The first is from domain expert acceptance test perspective which uses standard user acceptance criteria metrics to evaluate the prototype. We found the prototype is accepted by domain experts on average 84% which is promising result compared with other similar systems. The other standard metrics used to measure retrieval performance is recall and precision. We acquired 88% of relevant cases are retrieved out of the total assigned relevant cases by domain experts which is a promising recall result. The rules generated using data mining techniques are in support of the specific case experienced knowledge. Such a clinical decision support systems are very significant in countries like Ethiopia where shortage of health professionals are high.