Mineral Prospectivity Mapping (MPM) is critical for predicting potential high mineral locations based on a thorough examination of complex geological data. Although fundamental, traditional approaches frequently face difficulties when dealing with high-dimensional data and class imbalance, which are common within the field of mineral exploration. As a result, advanced machine learning methods have been used to improve the accuracy and efficacy of MPM, most notably Contrastive Representation Learning (CRL) and Twin Learning for Dimensionality Reduction (TLDR). CRL is an excellent tool for processing raster data because it can accurately identify potential mineral deposits by converting them into a lower-dimensional space while preserving important information. Combining CRL with TLDR, which effectively reduces redundancy while maintaining the local structure of the data, creates a robust framework for managing large geospatial datasets. The study shows that CRL works better than other dimensionality reduction methods, especially when used on raster data. It also fills in important gaps in mineral prospectivity mapping, like the problem of class imbalance. The proposed method enhances prediction accuracy and enables a clear spatial representation of mineral sites, becoming a significant tool for future geoscientific studies.