Breast cancer remains one of the leading health challenges affecting women worldwide, where timely and accurate prediction is essential for improving diagnosis and treatment outcomes. Among various risk indicators, body mass index (BMI) has been recognized as a significant contributor to breast cancer risk estimation. In this work, a multi-task classification approach based on the directed multitask twin support vector machine (DMTSVM) is introduced to enhance breast cancer prediction using BMI-related features. To address population diversity, BMI data are partitioned into two distinct groups, each modeled as an individual task within the multitask learning framework. The proposed model incorporates two key balancing parameters, whose optimal values are determined using two metaheuristic optimization techniques: the Marine Predators Algorithm (MPA) and the Grey Wolf Optimizer (GWO). By integrating these optimization strategies, the predictive capability of the DMTSVM model is significantly improved. Experimental results obtained from a breast cancer dataset structured around BMI-based tasks indicate that the optimized DMTSVM consistently outperforms its non-optimized counterpart. These findings highlight the potential of combining multitask learning with metaheuristic-based parameter optimization for effective medical prediction applications.

Keywords