PSO Based Fast K-means Algorithm for Feature Selection from High Dimensional Medical data set

Features are the most important entity in any data mining and machine learning applications. They are the backbone of any model. Reliability, efficiency and accuracy of the model depends upon the choice of strong and relevant features. However, feature selection is always a time-consuming and challenging task. In this paper, we have proposed an approach where we combine a clustering technique and a stochastic technique to select effective features from the high dimensional breast cancer data set in quick time. In order to select strong and relevant features, we have used an improved version of K-means algorithm called fast K-means algorithm, which is much faster and more accurate than a general means algorithm. The fast K-means algorithm is embedded in Particle Swarm Optimization (PSO) algorithm to produce better results. The results were validated using various classification techniques and were evaluated on various performance evaluation measures. The results obtained were found to be highly supportive in nature. The feature subset generated using PSO based fast K-means algorithm on KDDcup 2008 data set produced an accuracy of 99.39% and its time complexity was found to be O(log(k)).