<p>Dust susceptibility mapping (DSM) is essential for environmental risk management and mitigation; however, existing studies often suffer from limited accuracy due to suboptimal machine-learning models and inefficient hyperparameter tuning. Previous research has explored various tree-based ensemble learning methods and metaheuristic optimization techniques, but a comprehensive evaluation of hybrid approaches remains scarce. To address this gap, this study presents an innovative DSM framework by integrating the Light Gradient Boosting Machine (LightGBM) with two advanced metaheuristic algorithms: Particle Swarm Optimization (PSO) and Heap-based Optimizer (HBO). The research was conducted in Bushehr Province, Iran—an area frequently impacted by dust storms due to its climatic and geographic conditions. The dataset was constructed using remote sensing and Geographic Information System (GIS) techniques, incorporating multiple environmental and climatic factors. The effectiveness of three models—LightGBM, LightGBM-HBO, and LightGBM-PSO—was evaluated using the Area Under a Receiver Operating Characteristic Curve (AUC-ROC) index. The results indicate that the LightGBM-PSO model surpasses the LightGBM-HBO and baseline LightGBM models in predictive performance. Specifically, the AUC values for LightGBM-PSO, LightGBM-HBO, and LightGBM were 0.992, 0.988, and 0.972, respectively, demonstrating the superior predictive capability of the PSO-optimized model. Furthermore, statistical analysis confirms the significant improvements achieved by LightGBM-PSO, with a Z-statistic of 2.584 (<i>P</i> = 0.0098) compared to the baseline LightGBM model. These insights offer a valuable framework for improving dust risk assessment methodologies using remote sensing and GIS-based approaches and support policymakers in implementing more effective dust mitigation strategies in Bushehr Province and other dust-prone regions worldwide.</p>