Optimizing miRNA-Based Breast Cancer Subtyping with AHALA: A Multi-Stage Classification Approach

Abstract

This paper presents the Adaptive Hill Climbing Artificial Lemming Algorithm (AHALA), a novel optimization method designed to improve miRNA-based breast cancer subtyping and classification. By integrating Hill Climbing into the Artificial Lemming Algorithm, AHALA balances global exploration and local exploitation through adaptive boundary handling and dynamic step-size decay. Applied to breast cancer miRNA expression data, AHALA excelled in multi-classification tasks, achieving an accuracy of 95.74%, precision of 95.98%, recall of 95.74%, F1 score of 95.74%, and AUC of 0.9682. These results highlight its ability to distinguish between breast cancer subtypes, including Luminal A, Luminal B, HER2-enriched, and Basal-like. Feature selection powered by AHALA identified key miRNAs--hsa-miR-190b, hsa-miR-429, hsa-miR-505-3p, hsa-miR-3614-5p, and hsa-miR-935--linked to subtype differentiation. Additionally, AHALA optimized neural network hyperparameters, including hidden layer size, learning rate, and batch size, enhancing classification performance. These findings demonstrate AHALAs potential as a robust tool for miRNA classification, biomarker discovery, and machine learning optimization in breast cancer research, contributing to improved diagnostics and treatment strategies.