Leveraging Machine Learning for Effective Breast Cancer Diagnosis

Breast cancer is a prevalent global health concern, constituting 25% of female cancer cases. Early diagnosis through mammogram screening is effective, but limitations exist, particularly in dense breast cases. Machine Learning (ML) emerges as a promising tool for precise diagnosis. This study aims to identify optimal ML strategies, classifiers, and feature selection techniques for breast cancer diagnosis. This study analyses three breast cancer datasets, categorizing them by location, type (benign or malignant), and recurrence. We evaluate twenty-two classifiers across six ML strategies, taking into account accuracy, precision, recall, and ROC area metrics. We employ five feature selection techniques on 50% of the features. The results are promising for the adoption of ML in breast cancer diagnostics, with accuracy reaching higher than 93% for some applications. It is found that the HT and J48 classifiers from the Trees strategy and the NB classifier from the Bayesian strategy revealed promising results in the diagnostics and detection of breast cancer compared to other analyzed classifiers. Using feature reduction techniques in detecting the type of breast cancer (Benign/Malignant), the correlationAtriEval technique was found to have the highest performance, while the RellieffAttriEval technique has the highest performance when employed for feature reduction in detecting types of breast cancer (recursive/non-recursive).