Fire incidents modify soil characteristics, intensify heavy metal contamination, and threaten ecosystems and human health. This study examined the influence of fire on soil heavy metal contamination in the Kutupalong Rohingya refugee camp located in Cox's Bazar, Bangladesh. We collected 170 spatial soil samples from the area and applied machine learning (ML) models, feature importance, and bivariate local indicator of spatial association (LISA). Among them, the extreme gradient boosting (XGBoost) model showed superior performance in predicting Ni (0.79), Cd (0.81), and Pb (0.58), attaining significant R(2 )values, while the random forest model excelled in predicting Cr (0.83), and As (0.87). The actual contents of Cr, Ni, As, Cd, and Pb were 25.2, 81.6, 4.02, 1.10, and 11.8 mg/kg, respectively, whereas the best-modeled contents were 24.5, 82.3, 4.11, 1.09, and 11.9 mg/kg, respectively. The feature attribute analysis revealed correlations among Cr-Ni, Cr-Fe, Cr-Zn, Ni-Zn, As-Cr, As-Fe, As-Pb, Pb-Mg, and Pb-B, characterized by their respective correlation coefficients. The developed models indicated moderate potential ecological risks (50 < RI < 300) for heavy metals in soils, revealing a high level of Cd contamination (160 < E-i (r) < 320). The Moran's I indices for Cr, Ni, As, Cd, and Pb were 0.54, 0.73, 0.64, 0.92, and 0.66, respectively, exhibiting robust positive spatial autocorrelation and supporting the clustering of similar contamination levels. This study highlights the need for sustainable soil management practices in fireprone regions to reduce heavy metal pollution. The amalgamation of ML and geospatial methodologies provides a comprehensive framework for the surveillance and management of post-fire soil contamination in susceptible areas such as the Rohingya Camps in Bangladesh.