Enhancing the performance of Classification algorithms through Dataset tuning feature selection and Exception handling

Abstract

This thesis presents a spectrum of novel solutions for enhancing the newlineperformance of classification tasks in Knowledge Discovery Process by newlinetaking into account redundant and inconsistent data irrelevant and redundant newlinefeatures and exceptions while performing classification tasks equal newlineprobability and information gain that can arise often in real world datasets newlineThese characteristics are very often neglected by state of the art classification newlineAlgorithms The main focus of this thesis is Knowledge Discovery in newlineDatabases KDD Four different domains related to KDD namely Dataset newlineTuning Feature Selection Handling Exceptions in Classification algorithms newlineand Classification of Threatening E mail have been investigated newlineI n real world datasets lots of redundant and conflicting data exist newlinethat affect the performance of the classification algorithms They have to be newlineremoved to increase the efficiency and the accuracy of the classifiers Dataset newlinetuning is a pre processing method used to fine tune the tuples in a dataset for newlinenoise elimination Dataset tuning is done to remove redundant data and to newlinecorrect the conflicting data newlineFeature selection is a fundamental problem in data mining to select newlinerelevant features and cast away irrelevant and redundant features based on newlinesome evaluation criteria It is well known that correlated and irrelevant newlinefeatures may degrade the performance of the classification algorithms In this newlinethesis feature selection algorithms such as Bayes Feature Selector Class newlineAssociation Rule Information Gain Feature Selector and Bayes Theorem newlineInformation Gain Feature Selector are proposed newline newline