Automated Prediction of Breast Cancer: Performance Comparison Between Classical ML Algorithms and CNN

Breast cancer is a condition in which aberrant breast cells grow uncontrollably, forming tumors that are extremely fatal if left unattended or undiagnosed. It is currently extremely prevalent globally. In 2022, women in 257 countries out of 185 were affected by breast cancer as the prevailing cancer type. It has existed from historic times, and can be traced back to the Egyptians, around 1500 B.C. In terms of deaths due to cancer, breast cancer stands as the principal driver. It caused 670,000 deaths globally in 2022, among the 2.3 million diagnosed women [1]. This necessitates accurate and timely detection for improved treatment outcomes and reduced mortality rates. This paper evaluates traditional machine learning frameworks (RF, Logistic Regression, SVM, KNN) and CNNs for detection of breast cancer using the Wisconsin Diagnostic dataset. Models are compared using accuracy, precision, recall, F1 score, and confusion matrix. Findings demonstrate that results obtained from the CNN are excellent and can be compared with the Logistic Regression and SVM model in terms of overall outcome. The results obtained using Logistic Regression and SVM are equivalent. Thus, this paper highlights the importance of utilizing machine learning for better diagnostic consistency.