Abstract
Automated mineral identification in thin-section petrography remains challenging due to limited datasets capturing complete optical characteristics across crystallographic orientations. The Menoufia University Machine Learning Dataset for Minerals Classification 2025 (MUMDMC2025) provides 14,400 high-resolution photomicrographs of five mineral classes: Biotite, Hornblende, Plagioclase, Potassium-Feldspar, and Quartz from Egyptian Eastern Desert granite samples. Each mineral specimen was systematically imaged at 72 rotational positions (5° increments, 360° coverage) under both Plane Polarized Light (PPL) and Cross Polarized Light conditions (XPL), documenting complete anisotropic optical properties including pleochroism, birefringence, and extinction patterns. This comprehensive rotational imaging protocol addresses critical gaps in existing petrographic datasets by capturing orientation-dependent optical phenomena essential for reliable mineral classification. The balanced dataset contains 2,880 images per mineral class, enabling robust machine learning model development and evaluation. Validation demonstrates dataset utility with K-Nearest Neighbors, achieving high classification accuracy. The dataset supports the development of automated petrographic analysis systems, quantitative mineralogical research, and educational applications in optical mineralogy, providing researchers with comprehensive optical documentation necessary for advancing computer-vision approaches in geological sciences.