Upgrading the Performance of KNN and Naïve Bayes in Diabetes Detection with Genetic Algorithm for Feature Selection

Abstract

Data mining is a science that is used for finding models and association rules in huge data where other statistical analysis cannot do that. The medical science needs data mining for analyzing massive data and producing predictive models. The purpose of this research is to present a framework for creating, evaluation and exploitation of data mining models. In this study we have used the combination of artificial intelligent technique such as feature selection with k Nearest Neighbor (kNN) and Naïve Bayes approach to develop a predictive model which classifies the patient as healthy and diabetic [1][2]. The main purpose of feature subset selection is to reduce the number of features used in classification while maintaining the acceptable classification accuracy. Our proposed Genetic algorithm (GA) for feature selection approach improves the classification accuracy and uses fewer input features [3]. Some attributes in the dataset may not be useful for diagnosis and thus can be eliminated before learning. The goal of feature selecting is to find a least set of attributes so that the resulting probability distribution of the data classes is close to the original distribution obtained by all attributes [4]. Irrelevant, redundant, or noisy data can be removed by using Genetic Algorithm which in turn improves the mining performance such as predictive accuracy and result comprehensibility