Breast Cancer Recurrence Prediction Model Using Voting Technique

Abstract

In recent times, prediction modeling using machine learning is gaining widespread recognition, due to its ability to facilitate the detection of critical features from large datasets (Salama and Abdelhalim Int J Comput Inf Technol 01:2277–0764, 2012). The areas of application are fairly diverse. The conventional approach involves several data mining classifiers individually to build knowledge discovery systems. The emphasis has shifted now for deploying an ensemble of these classifiers in order to further enhance the prediction capabilities and accuracy of the knowledge discovery database (KDD) process. This current study proposes a model for prediction of the recurrence of breast cancer within 3 years, based on one of the types of ensemble data mining classification techniques called voting. This approach uses different combinations of four data mining base classifiers, viz., decision tree, multilayer perceptron, Naïve Bayes, and SMO. An attempt is being made to compare the effectiveness of voting classifiers, vis-a-vis the base classifiers in order to determine the performance-enhancing capabilities of the ensemble approach. Our work clearly demonstrates that the performance accuracy of the voting classifiers analyzed with seven combinations is consistently high with values ranging between 81.0526% and 83.8596%. In contrast, the performance accuracy of base classifiers varies widely ranging between 75.7895 and 84.2105%. We have clearly established that the performance of the voting classifier is very consistent. Voting also enhances the performance of weak classifiers like MLP and SMO. The dataset used in our experiment consists of 23 attributes containing 575 samples obtained from the Mizoram Cancer Institute of Aizawl, Mizoram, India.