Experimental and Statistical Analysis on the Performance of Firefly based Predictive Association Rule Classifier for Health Care Data Diagnosis

Abstract

Health care data diagnosis deals with a prediction of the course of a disease by analyzing the information in health care systems. Analyzing healthcare datasets is one of the major challenges of recent times. Associative Classification (AC) is one of the data mining techniques commonly used for disease diagnosis. AC integrates the concept of Association Rule Mining (ARM) and classification. Though, AC is an efficient classification system, it often experiences poor accuracy as it generates huge volume of class rules in the ‘rule generation’ phase. This paper intends to address this issue by structuring an associative classifier using significant PARs (Predictive Association Rules) i.e. simply class rules. In this work, Firefly Algorithm (FA), a nature inspired metaheuristic optimization algorithm is adopted to fit into the ‘rule generation’ phase of existing CPAR (Classification based on Predictive Association Rule), an AC algorithm. This work acquires the essential inspiration of FA and CPAR to construct an associative classifier with significant PARs. FA with a customized fitness function specifically designed for the health care data diagnosis is proposed to find a small set of significant PARs. FA based Predictive Association Rule (FPAR) classifier thus built using significant PARs achieves high prognostic accurateness and interestingness value. Performance of FPAR and CPAR algorithms are analyzed over the six health care datasets from UCI machine learning repository. Based on the experiments, promising results in terms of classifier accuracy are provided by FPAR algorithm.