ANALYSIS OF OPEN COAL MINING POTENTIAL AREAS BASED ON SATELLITE IMAGE DATA AND MACHINE LEARNING

• Main Author: Bima Makmunar Syamsi, Arya
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73435
• Institution: Institut Teknologi Bandung

<p align="justify">This research aims to explore the potential use of machine learning in coal exploration in Indonesia using satellite image and drillhole data. Coal is one of the largest energy sources in Indonesia and has significant deposit reserves. However, open-pit mining prospects become increasingly difficult in the future due to coal layers being located deeper than the surface, resulting in a higher ratio of coal to rock and reaching uneconomical values. Therefore, determining areas with coal potential requires interdisciplinary exploration methods to identify coal deposit potential. This study used drillhole data as training points and testing for machine learning model development to classify potential coal mines. Random Forest and CART machine learning models were used with 60:40, 70:30, 80:20, and 90:10 data training and testing splits. The goal of using machine learning models is to optimize coal potential analysis in a location. The research results are expected to provide useful information for efficient coal exploration in the future. The research results show that the study area has a coal volume of 293524743 m3. In addition, the best classification accuracy was obtained by Random Forest 80:20 and CART 90:10, while Cart 80:20 was the only one that could identify true positives in validation data with the largest variable importance in band-4 at 0.543 and accuracy of 0.777778, precision of 1, recall of 0.333333, and f-1 score of 0.5. The correlation between data was not significant.