Enhanced Multi-Objective Optimization Algorithm Using Adaptive Fitness Function and Penalty

Multi-objective optimization (MOO) plays a critical role in real-world decision-making systems where multiple conflicting objectives such as maximizing classification accuracy, minimizing validation loss, and preserving model generalization must be optimized simultaneously. Traditional MOO techniques, including NSGA-III, rely on static fitness functions and fixed penalties, which limit adaptability and often result in premature convergence and reduced solution diversity. This paper proposes an enhanced NSGA-III framework that integrates dynamic fitness evaluation and adaptive penalty weighting guided by population evolution and dataset sensitivity. The proposed method is evaluated on 11 real-world medical datasets, demonstrating improved convergence, diversity, and robustness compared to conventional NSGA-III and baseline models.