Optimization-Based Feature Selection and Ensemble Machine Learning Algorithms for Breast Cancer Classification

Abstract

Breast cancer, which originates in a woman's breast tissue, is acknowledged to be a significant study topic in the medical field. For a long time, there has been a serious concern with the classification of breast cancer. Thus, to effectively categorize the breast cancer dataset, machine learning methods are designed and implemented. In previous research, the algorithms have classification accuracy and time complexity issues. This study proposes the use of Enhanced Cuckoo Search Optimization combined with Ensemble Machine Learning Classifiers (EMLC) to tackle the identified challenges and improve the accuracy of breast cancer classification. The system is structured into four key stages: pre-processing, feature extraction, feature selection, and classification. During pre-processing, statistical correlation analysis is applied to eliminate noise from the dataset, thereby enhancing classification performance. The feature extraction phase then derives more informative features from the cleaned data to support more accurate classification. It is performed using Improved Principal Component Analysis (IPCA), which extracts the prominent features from the breast cancer dataset. Then, utilizing the best fitness values of cuckoos, the ECSO algorithm is utilized to identify the relevant and useful characteristics. Finally, using a training and testing model, the EMLC algorithm is employed for classification. It classifies the features more accurately using ensemble Enhanced Granular Neural Network (E-GNN), Adaptive Neural Fuzzy Inference System (ANFIS) and Weighted Support Vector Machine (WSVM) algorithms. The experimental findings show that the proposed EMLC algorithm achieves superior performance compared to existing approaches, offering improved precision, recall, F-measure, accuracy, ROC curve results, AUC scores, and lower time complexity.