The Philippine government emphasizes responsible mineral development, advocating for sustainable exploration and conservation of mineral resources [1]. Lithologic mapping plays a vital role in identifying rock types associated with valuable minerals like nickel, gold, and copper. Traditional lithologic mapping methods, however, are labor-intensive, costly, and time-consuming, especially in remote or inaccessible areas [2]. This study evaluates a machine learning approach utilizing Sentinel-2 multispectral data to map lithologic units in Mayantoc, Tarlac, and San Juan, Batangas, regions with significant economic potential. The study employs a Multilayer Perceptron Artificial Neural Network with Backpropagation (MLP-ANN-BP) as the learning algorithm to enhance spatial and spectral classification accuracy. The results demonstrate the model’s robustness, even with low sample sizes. At 250 sample points per lithologic unit, the model achieved an accuracy of 80.72%, which improved to 89.29% with 1000 points. Notably, even at smaller sample sizes, the model showed fair performance, with certain classes. Hyperparameter tuning further optimized performance, raising accuracy to 81.95% for 250 points and enhancing classification metrics across most classes. While challenges remained in distinguishing specific lithologic units due to feature overlap, the model proved effective in delivering reliable predictions with limited data. This study underscores the potential of machine learning-based lithologic mapping as a cost-effective and scalable solution for sustainable mineral resource exploration in the Philippines, particularly in regions characterized by vast and geologically diverse landscapes.