Particle Swarm Optimization Feature Selection for Breast Cancer Prediction

Abstract

One of the most common diseases faced by women is breast cancer. Early diagnosis is very important to prevent cancer from getting worse. The conventional diagnosis process needs more time. The use of machine learning methods to diagnose diseases can be a quick and practical solution. However, advances technology produces various types of high dimensional data, especially relating to medical or cancer data. High dimensionality of the data makes it more difficult to get insights from them. Unrepresentative data can lead to biased classification results. The feature selection method can be used to overcome some of these problems. In this study, we use Particle Swarm Optimization (PSO) as a feature selection to improve the performance of classification algorithm, namely, SVM, Naive Bayes, Logistic Regression, Decision Tree, and KNN. We use that classification algorithm to evaluate the fitness size and as classifier. We also compare PSO performance with a renewed algorithm, the Genetic Algorithm. Using data from UCI Breast Cancer Dataset proves that Particle Swarm Optimization can improve several classification algorithm performances. However, PSO can't overcome performance of Genetic Algorithm as feature selection.