Groundwater serves as a crucial resource for sustaining agriculture, providing drinking water, and supporting industrial processes. This research aims to develop an improved XGBoost-based machine model for groundwater level prediction using soil properties and resistivity data. By optimizing XGBoost Classifier and SHAP (SHapley Additive Explanations), the study seeks to enhance the accuracy and interpretability of groundwater detection methods. The dataset, collected through resistivity surveys, includes critical attributes such as resistivity, depth, and conductivity. The proposed methodology involves data pre-processing, model training, and SHAP analysis to identify key features influencing the presence of groundwater. The findings highlight the efficacy of the XGBoost model in predicting groundwater levels, supported by a 2D subsurface visualization tool for practical application in hydrogeological studies. SHAP's interpretability feature played a crucial role in explaining the model's decision-making process, ensuring the results are transparent and trustworthy. This integrated approach significantly improves the reliability of groundwater prediction systems, making it valuable for environmental and water resource management.