A Comparative Analysis of Clustering Techniques with Feature Selection for Predicting Breast Cancer Recurrence

After recovery from breast cancer, chances are there for the recurrence of cancer where it comes back after initial treatments. Conventional methods to detect breast cancer(BC) recurrence such as watch for recurrent cancer during the followup examinations are done by oncologists take a longer time to reveal the threat. Thereby, early prediction and detection of recurrence cancer is the vital need of every patient since it is their greatest fear. Since prediction has to be fast and efficient, highly influenced features for the recurrence should be identified. This research aims to comparatively analyze the advantages of using feature selection on clustering algorithms in predicting BC recurrence. In this study, we have performed breast cancer recurrence prediction using clustering techniques namely, Kmeans algorithm, Complete-Linkage(CL) algorithm, Expectation-Maximization(EM) algorithm, and Hierarchical Kmeans(HK) algorithm and Information GainfIG) as the feature selection. We studied the outputs from the clustering algorithms with IG feature selection and compared them with outputs from clustering algorithms without using feature selection. Finally, the results exhibited that the EM algorithm with IG feature selection outperformed the other clustering techniques.