<p>In the past few years, the use of Machine learning (ML) to classify remotely sensed data has increased, offering new opportunities for geological mapping. Conventional remote sensing classification methods often rely on spectral information, but distinguishing between lithological classes with similar spectral signatures remains a persistent challenge. In particular, accurately mapping and extracting pegmatites from other lithological classes, especially granite, presents a difficulty. The objectives of this study are to map the lithological units in the Angarf region (Zenaga, Central Anti-Atlas, Morocco) and to extract pegmatite outcrops, with a particular focus on separating the pegmatite from the granite, as this challenge has been considered in several previous studies. The methodology developed is innovative and based on a Decision Tree (DT) approach of ML, which is applied to spectral indices derived from ASTER (Advanced Space borne Thermal Emission and Reflection Radiometer) images. The interpretation and analysis of spectroradiometric measurements have enabled us to understand the behavior of spectral information of pegmatites compared to other geological formations. The achieved overall accuracy of the DT classification was 96.28 %. Also, the comparison of the produced map, particularly the pegmatite classes, with the field data highlighted the potential of the adapted methodology. The DT algorithm is a fast, reliable, robust, and highly accurate mapping model that is simple to configure, uses few parameters, and handles input data heterogeneity effectively. The obtained pegmatite maps provide a support and can be used as a preliminary step in mineral exploration.</p>