A Hybrid Machine Learning Framework with Metaheuristic Feature Selection for Robust Breast Cancer Detection

Abstract

Breast cancer is one of the most prevalent malignancies worldwide, posing a significant health threat, particularly to women. Early detection is crucial for improving survival rates, yet conventional diagnostic methods often suffer from inaccuracies and delays. Advances in artificial intelligence and machine learning have introduced promising solutions for early and precise detection. This study proposes an advanced multimodal machine learning system integrating radiological and genomic data for improved breast cancer detection and prognosis. The Wisconsin Diagnostic Breast Cancer (WDBC) dataset, comprising 569 samples with 30 numerical features representing tumor characteristics, is utilized for training and evaluation. A novel real-time hybrid machine learning approach combining Multi-Layer Perceptron (MLP) and Logistic Regression (LR) is developed to enhance classification accuracy while ensuring model stability and clinical applicability. Feature extraction techniques, including Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), are employed to select the most relevant attributes, optimizing predictive performance and reducing computational complexity. Comparative analysis highlights the superiority of machine learning over traditional diagnostic methods. The ensemble model achieves an accuracy of 96.49%, recall of 99%, and precision of 98%, demonstrating its robustness in real-world applications.