Indian Journal of Science and Technology

Abstract

Objective: Even though there are many advanced and sophisticated data mining techniques that are being used to enhance the quality of health services, many of them still fail to produce accurate predictions when applied to real-time health datasets. Therefore, our aim is to devise the most accurate prediction model for an early diagnosis of diseases based on previously recorded patient data without performing any extensive laboratory tests. Methods/Statistical Analysis and Finding: Datasets are discretized by converting numerical attributes into categorical attributes. To such datasets, a decision forest algorithm is applied to produce a diverse group of classifiers. The algorithm eliminates the appearance of identical attributes in subsequent trees. Evaluation of a proposed hybrid model showed that the new technique had successfully improved the accuracy of classification and prediction by 9~10%. The accuracy is increased by reducing the information loss using a low-frequency discretization technique and by enhancing classification capabilities by generating a diverse group of classifiers. Application/Improvement: The proposed hybrid model combines two advanced techniques Lowfrequency Discretization to reduce information loss during attribute discretization and to increase diversity and Forest PA to increase the accuracy of classifiers

Keywords: Data Mining, Data Visualization, Early Diagnosis of Diabetes, Forest PA, Knowledge Discovery, LFD, Sophisticated Data Mining Techniques