Dust storms pose significant environmental and health challenges in arid regions, necessitating accurate modeling for effective mitigation strategies. This study employed thirteen machine learning (MLs) and deep learning (DLs) models to identify dust-prone areas and evaluate the impact of various environmental drivers on dust storms. A set of variables, including evapotranspiration (ET), air temperature, land surface temperature, vegetation indices, mean precipitation (Pr), soil moisture, and six drought indices (Standardized Precipitation Index (SPI), Vegetation Health Index (VHI), Vegetation Condition Index (VCI), Temperature Condition Index (TCI), Soil Vegetation Index (SVI), and Palmer Drought Severity Index (PDSI)), was analyzed using Aerosol Optical Depth (AOD) as the target variable. The analysis revealed that tree-based MLs outperformed DLs in this study area, potentially due to the regional scale and dataset characteristics. Random Forest (RF) emerged as the most outstanding model, achieving exceptional accuracy in both regression (R-2 > 0.96, RMSE = 0.01, MAE = 0.01) and classification tasks (Critical Success Index = 0.70, Recall = 0.76), along with a Bias value of 1.038 and 85 % overall accuracy in spatial detection of dust sources. Among DLs, Artificial Neural Network (ANN) showed competitive performance as a reliable alternative. Variable importance analysis identified temperature, precipitation, and ET as the most influential predictors, followed by soil moisture and PDSI. The findings provide a good framework for dust susceptibility mapping and highlight the advantage of tree-based MLs for dust modeling in regional-scale studies.