Particle Swarm Optimization for Feature Selection to Improve Performance of Decision Tree Classifiers

Abstract

In many applications, people are dealing with massive data that contains newlinemultidimensional attributes such as network intrusion data, stock market data, medical newlinedata, weather forecast data and much more. Thus, data classification is faced with a newlineproblem when it has to generate rules with many attributes or features. The time newlinerequired to generate rules is proportional to the number of features. In addition, newlineirrelevant and redundant features can reduce both the predictive accuracy and newlinecomprehensibility of the induced rule and degrade the classifier speed (due to its high newlinedimensionality). Thus, selecting the most relevant features is necessary, and this newlinestrategy is implemented to simplify the rules and reduce its computational time while newlineretaining the quality of classification, as it represents the original features set. In this newlinework we attempt to implement an algorithm that investigate the accuracy of decision newlinetree techniques combine with PSO on medical image dataset. newlineFirstly, we start by preprocessing the medical images, in order to remove the newlinedigitization noise using Wiener filter and enhance the contrast of breast tissue using newlinemodified fuzzy pal and king method. Next, K-mean clustering method is used to newlineidentify the ROI and from that ROI 22 features are extracted using GLCM technique. newlineIn the next phase BPSO is used to search for the feature subset and K-nearest newlineneighbor (KNN) classifier is used to evaluate the feature subset. A decision tree newlineclassifier is then trained using the significant features in the training set found by newlineBPSO-KNN feature selection method the trained decision tree classifier is then used newlineto classify the ROIs in the test set. For the purpose of experiments we used mini newlineMAIS dataset. Experimental result shows that proposed method achieves higher newlineaccuracy. newlineProposed algorithm is implemented in MATLAB 2020 Experimental results show that newlineaverage classification accuracy of decision tree classifier without and with BPSOKNN newlinefeature selection method are 85.35 and 87.96 respectively. On the basis of newlinean