Predictive modeling of mineral prospectivity by deep self-organizing map: Implications for copper exploration targeting in saveh district, central Iran

• Combining unsupervised deep learning (DAN) for dimensionality reduction with SOM’s clustering to delineate ore-related spatial patterns. • Validating results through multi-source geospatial data from the Saveh district (Central Iran), P-A plot analysis , and field evidence ( petrography, microprobe data ). • Demonstrating superior targeting accuracy compared to conventional methods, with direct implications for cost-effective copper exploration . The increasing challenges in mineral prospectivity mapping (MPM), particularly for deep-seated and complex metal deposits, demand the application of cutting-edge computational workflows to model mineralization-related spatial patterns effectively. This study discusses a novel hybrid algorithm combining a deep autoencoder network (DAN) with a self-organizing map (SOM) to recognize mineralization-related spatial patterns and enhance the MPM procedure. By integrating the dimensionality reduction power of the DAN with the robust clustering capabilities of the SOM, the proposed workflow aims to achieve more precise prospectivity predictions. The proposed hybrid algorithm was applied to multi-source exploration data pertaining to the Saveh district, Central Iran. A comparative analysis with traditional approaches was constituted, whereby prediction-area (P-A) plots were used to assess the relevance of models in predictive modeling of mineral prospectivity within the study area. The results derived from the P-A plots demonstrated the superiority of the proposed hybrid model in delimiting target areas for further metal exploration. Moreover, field surveys and mineralogical studies, including petrographic and microprobe analyses, confirmed the presence of ore-bearing evidence within the identified target areas. The findings suggest that the proposed methodology has the potential to enhance MPM accuracy and mitigate costs and risks in regional-scale exploration programs. The research demonstrates the potential of integrated deep learning and clustering techniques in boosting MPM procedures and discovering new metal deposits within complex metallogenic systems.