PREDICTING PHYSICAL AND MENTAL HEALTH RISK IN ELDERLY POPULATIONS: A MACHINE LEARNING APPROACH WITH STRUCTURED FEATURE ENGINEERING

This research introduces a systematic methodology for multi-class health risk prediction in aging populations, addressing issues of unequal class distribution and inaccuracies in tabular healthcare data. This action was undertaken to resolve issues. The raw data undergo pre-processing and feature selection before to analysis. This technology is often upgraded to improve health metrics. The data is subsequently classified into three risk categories: low, moderate, and high rates, utilizing clinically informed adjustments. This is executed to guarantee that the predictive task aligns with medical interpretation rather than merely statistical categorization. The system employs a Chaotically Guided Hybrid Swarm Optimizer with Adaptive Echo Refinement (CHSO-AER) to surmount conventional optimization limitations. This approach introduces controlled randomness at the beginning of the search, which is modified during the iteration process. This is done to prevent premature convergence and to offer stability in between iterations. Since ensemble-based models may be extremely sensitive to even little changes, this function is particularly important for fine-tuning sensitive parameters. This is because models may be extremely sensitive to little changes. For this particular reason, this functionality is quite significant. Three gradient boosting techniques are employed in a layered meta-learning approach to generate predictions. The ability of these algorithms to improve accuracy led to their selection. Thus, the ensemble is more consistent across data subsets. Experimental assessment uses a 1,200 record cognitive health dataset and a 100,000 sample physical health dataset. Performance improvements are observed gradually rather than all at once, demonstrating the influence of every pipeline component. The results of the proposed optimizer show more stability and dependability than those of GA, PSO, and GWO. The proposed model achieves an accuracy of 97.0% and an F1-score of 0.966 on the physical dataset. With an F1-score of 0.968, it yields equivalent results on the cognitive dataset. Prediction errors are limited and fairly equal, as further evidenced by class-specific performance. The entire method makes generalization easier, particularly for structured medical classification issues where reliability and multi-class prediction are crucial.