Comparative evaluation of automated machine learning techniques for breast cancer diagnosis

Abstract

Breast cancer is the second leading cause of death among women worldwide. Early detection is crucial for a high possibility of recovery, but current diagnostic procedures rely on visual inspection by doctors, which is time-consuming, requires consultation, and lacks reliable automatic detection systems. Therefore, an automatic diagnosis system based on machine learning (ML) models is highly required to improve the accuracy of detection and prediction. In this study, an efficient classification method based on ML techniques was proposed to help doctors diagnose and distinguish between malignant and benign tumors relevant to breast cancer prediction. Seven AutoML techniques, including Orange, Lazy Predict, TPOT, MLJAR, MATLAB classification learner, and AutoKeras, were examined and applied to eight different datasets to evaluate their performance in terms of classification accuracy. The experimental results demonstrate a significant improvement using state-of-the-art techniques, with Lazy Predict and MATLAB Classification Learner outperforming all other ML techniques for binary classification tasks. The contribution of this work is the identification of the best ML model among multiple models inside each AutoML technique applied to different breast cancer datasets. All datasets and codes used in this study are freely available from https://github.com/amrrashed/breast-cancer-datasets-and-codes.