Strategic Dataset Splitting for Improved Breast Cancer Classification in Logistic Regression Based CAD System

Abstract

Breast cancer (BC) remains a prevalent global health concern, necessitating advanced diagnostic tools for early detection and accurate classification. This article introduces a Computer-Aided Diagnosis (CAD) system, leveraging a logistic regression model optimized with grid search, to classify BC using the Wisconsin Breast Cancer Diagnostic (WBCD) dataset. The study focuses on the impact of an optimal dataset-splitting strategy, proposing a ratio of 83:17 based on the count of unique rows in the input matrix, in contrast to a conventional 70:30 split. The CAD system, adhering to the optimal splitting ratio, demonstrates superior performance across various metrics compared to the 70:30 ratio. The findings underscore the potential effectiveness of strategic dataset splitting based on data quantity, highlighting its significance in enhancing the logistic regression model's classification accuracy.