<![CDATA[The digital transformation of higher education has generated large volumes of student activity data, most of which are stored as unstructured text. These data have significant potential to provide insights into learning patterns, competencies, and academic potential; however, conventional approaches often fail to capture contextual meaning and semantic relationships within student activity records. This study proposes LEARN MAP (Learning Activity and Academic Potential Mapping), an artificial intelligence framework based on Natural Language Processing (NLP) and Bidirectional Encoder Representations from Transformers (BERT) to automatically classify student learning activities and map academic potential. The study utilized a dataset of 8,954 student activity records collected from learning and Independent Learning–Independent Campus (MBKM) programs. The research process consisted of data collection, text preprocessing, BERT fine-tuning, activity classification, soft-skill mapping, and academic potential analysis. The results demonstrate that the BERT model achieved an accuracy of 92.31%, precision of 91.27%, recall of 90.84%, and F1-score of 91.05% in classifying student activities. The findings indicate that student activities were predominantly associated with communication, community engagement, collaboration, and research. Furthermore, the soft-skill mapping revealed that communication and self-directed learning were the most developed competencies, while most students were categorized as having moderate to high academic potential. The novelty of this study lies in the integration of BERT-based activity classification with soft-skill and academic potential mapping within a unified analytical framework. LEARN MAP has the potential to support data-driven academic decision-making and advance digital transformation initiatives in higher education.]]>
