Hyperspectral (HS) sensors are widely used for geological surveys and mineral classification. In areas with minimal vegetation and exposed minerals, it is ideal for mineral maps to remain consistent regardless of imaging time or sensor used. However, practical mineral maps often vary due to factors like atmospheric correction, sensor calibration, mixed pixels, and noise. This study introduces an approach combining the classification methodology of the knowledge-based expert system USGS Tetracorder with a data-driven approach, aiming to minimize discrepancies between mineral maps derived from independent datasets. To reduce the influence of noise and mixed pixels in HS data, Minimum Noise Fraction (MNF) transformations were applied, followed by a Pixel Purity Index (PPI) analysis. High-entropy pixels identified through PPI analysis were found to exhibit distinctive mineral-specific features compared to other pixels. These high-purity pixels were processed using the conventional Tetracorder method, serving as ground truth. For pixels deemed pure by PPI analysis, Tetracorder-derived mineral labels were used as the target variable, while the MNF-transformed bands were employed as feature variables. A Random Forest classifier was trained on these data and subsequently applied to map minerals in the remaining pixels. This method utilized the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) as the HS sensor and was validated over overlapping regions in the Cuprite area of Nevada, USA, using datasets captured at different times during the years 2011 to 2013. The results indicate that the proposed method provides a comparable level of accuracy to the standard Tetracorder implementation while significantly improving robustness and reducing errors compared to conventional processing. Future work will focus on validating this method in other regions and with different sensors to evaluate its generalizability.