Comparison of different machine learning methods on Wisconsin dataset

Abstract

Breast cancer research has made a great progress in the recent years, but there is still a room for an improvement. Wisconsin Diagnosis Breast Cancer (WDBC) contains 569 patients records with 32 attributes extracted from the digitized images of a fine needle aspirate of a breast mass. We used this dataset to compare selected machine learning methods in the binary classification solution. We realized the whole analytical process in accordance with the CRISP-DM methodology representing one of the most used process models for this purpose. Finally, we compared our results with some of the previously published research papers to evaluate our approach and expectations. We achieved the best accuracy with SVM - 97.66%, Random Forests - 97.37% and C4.5 - 95.61%.