Improving Open-Pit Mining Mapping Accuracy in the Tropics Using Enhanced Input Selection for Classification Process of Machine Learning

Indonesia possesses many open-pit mining that require monitoring. This study aims to map open pit mining in Central Bangka Regency, Bangka Belitung Islands Province, Indonesia using multi-temporal Sentinel-2 imageries. The open pit mining was mapped using Enhanced Input Selection for Classification Process (EISCP) and Machine Learning. The EISCP systemically integrates band selection, texture analysis employing the Gray Level Co-Occurrence Matrix, and Principal Component Analysis. Multiple Machine Learning (ML) algorithms, including Random Forest (RF), Classification And Regression Trees Classifier (CART), and Support Vector Machine (SVM), were utilized to classify mining and non-mining areas. There were three band combination scenarios: (1) a combination of blue, green, red, red edge-1, red edge-2, red edge-3, Near-InfraRed (NIR) bands, red edge-4, Short-Wave InfraRed-1 (SWIR-1), and Short-Wave InfraRed-2 (SWIR-2); (2) a combination of red bands, red edge-1, and SWIR-2; and (3) using principal analysis (PC1) from the EISCP results as classification data input. High resolution image of SPOT-6 and PlanetScope data in 2021 was used as data reference for validation. We found that the SVM algorithm with EISCP input band scenarios produced the highest accuracy of 97.55% and a kappa coefficient of 0.91 among all combinations of band scenarios and ML algorithms. The implementation of RF and CART algorithms with EISCP input band scenarios resulted good accuracy, with differences of only 0.62% and 1.84% from SVM. The mapping outcomes from the ML algorithms demonstrated that utilizing EISCP can enhance the precision of open-pit mining mapping in the research area.