Poverty elimination is an essential and unavoidable step in human development. Predicting poverty is the first crucial step in addressing this issue, especially in developing countries where it remains a significant concern. The main objectives of this paper are: (i) to explore and analyze the multidimensional poverty data in Pattani province, and (ii) to apply feature selection techniques (Chi-Square, Mutual Information, and Gini Index) to enhance the prediction process. These techniques help in reducing irrelevant and redundant features, leading to more efficient models and better insights. This paper presents the development of a predictive model aimed at classifying poverty and providing actionable recommendations for policymakers. Machine learning models, including Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM), were employed to assess the impact of feature selection methods on model performance. The effectiveness of each model was evaluated through various metrics, including accuracy. The experimental results show that the success of these models in predicting poverty, with DT, RF, and SVM obtaining 93%, 95%, 94% of accuracy, respectively. The findings underline the importance of feature selection in improving the performance of machine learning models.