Mineral prospectivity mapping, as a key tool in mineral exploration, faces challenges such as complex nonlinear relationships, data heterogeneity, and the limitations of labeled datasets. In this study, an unsupervised hybrid deep learning approach based on a Long Short-Term Memory (LSTM) network is proposed to identify prospective mineralization zones. The proposed model is designed to hidden structures, similarities, and anomalies in geoscientific data without requiring labeled training samples. Initially, the dataset is automatically grouped using statistical and algorithmic criteria, after which targets with the highest mineral prospect are delineated using a confidence index. To evaluate the effectiveness of the proposed approach, the model was applied to the 1:100,000 Kajan sheet in Isfahan Province, Iran, an area with significant Cu–Au mineralization potential. The input data include alteration, geochemical, fault, and intrusive rock, which were transformed using fuzzy logic prior to being fed into the deep learning framework. The results demonstrate that the proposed model is capable of capturing complex patterns and effectively delineating prospective zones, with outputs showing strong spatial consistency with known geological features.

Compared to conventional deep learning techniques, the proposed method offers notable advantages by eliminating the need for labeled data and enhancing the modeling of nonlinear relationships. Overall, the findings suggest that unsupervised deep learning models, particularly those based on LSTM architectures, provide a robust and efficient framework for mineral prospectivity mapping and can significantly support decision-making in mineral exploration