An improved extreme learning machine algorithm for prospectivity mapping of copper deposits using multi-source remote sensing data: a case study in the North Altyn Tagh, Xinjiang, China

Traditional extreme learning machine (ELM) model suffers from instability due to random initialization of input weights and hidden-layer bias, often resulting in suboptimal predictive performance. To address this limitation, the Slime Mould Algorithm (SMA), a bio-inspired optimization strategy, was integrated to refine the initialization of ELM parameters for mapping prospectivity of copper deposits. Based on 15 known ore deposits in the northern part of the Altyn Tagh, China, multi-source remote sensing data, including ASTER, Landsat-8 OLI, and ZY-1 02E imagery as well as geological data were integrated within a geographic information system (GIS) framework. The confusion matrix, the area under the receiver operating characteristic (ROC) curve and the success-rate curve were employed to evaluate the model’s performance against traditional ELM and multi-layer perceptron. Experimental results demonstrated that the SMA-ELM model exhibited superior overall performance compared to the other two models. Based on the Youden index, high-potential metallogenic zones were delineated which were consistent with known deposits and geological structures. These findings validate SMA-ELM as an effective and promising tool for mineral prospectivity mapping in geologically complex terrains.