<p>The scheduled caste (SC) and scheduled tribe (ST) populations in Dakshin Dinajpur, West Bengal, face significant socio-economic challenges, reflecting the broader issue of multidimensional poverty in marginalized communities. This study aims to quantitatively assess the deprivation experienced by these groups across multiple indicators using the multidimensional poverty index (MPI). The objectives include identifying distinct poverty groups through clustering and determining the key factors contributing to different poverty magnitudes using machine learning algorithms, thereby enabling the formulation of targeted intervention strategies. Using household survey data, the multidimensional poverty index (MPI) was calculated across health, education, and living standards dimensions. Cluster analysis revealed distinct groups based on MPI values, highlighting varying degrees of deprivation among SC and ST households. K-means clustering was applied to categorize households into different poverty groups based on their MPI values, enabling a more granular analysis of poverty distribution. Machine learning models, including random forest (RF) and gradient boosting machines (GBM), were then employed to identify key factors driving poverty within each cluster. The results show that ST households exhibit a higher MPI (0.321) than SC households (0.205), with factors such as education (38.83% for SC, 34.15% for ST) and nutrition (31.09% for SC, 32.28% for ST) being the most significant contributors. GBM outperformed RF in predicting poverty levels within each cluster, achieving R² values close to 1.0. The findings highlight the severe deprivation faced by ST households, particularly in education and nutrition, compared to SC households. These findings underscore the need for tailored poverty alleviation strategies, focusing on cluster-specific drivers such as education, income generation, and improved access to basic services, rather than applying uniform policies across all groups.</p>