Abstract
Feature selection for medical diagnostic models must account for feature interdependencies rather than independence. Ant colony optimization (ACO), a heuristic algorithm inspired by ant foraging, effectively addresses feature selection and correlation challenges. Published studies show that integrating ACO with other algorithms and refining pheromone-heuristic computations significantly boosts medical data feature selection performance. To tackle the issue of features correlation in medical diagnostic models, this study presents the L-S-ACO algorithm by combining the ACO algorithm with two additional algorithms: Light gradient boosting machine (LightGBM) and stochastic average gradient (SAG). Based on this algorithm, a medical diagnostic model is developed, and its practical application in disease diagnosis is analyzed. The analysis results demonstrate that the proposed algorithm improves accuracy (ACC) by 5.81%, F1 score by 6.21%, area under the curve (AUC) by 4.08%, precision by 5.12%, and recall by 6.35%.