AI-driven competency recommendations based on attendance patterns and academic performance

This study introduces an Artificial Intelligence (AI) framework that analyzes attendance patterns and learning outcomes to generate personalized competency recommendations (CR). Data from several Indonesian universities covered 22,304 students across 46 courses mapped to 90 competencies. Gradient Boosting (GB) was the most effective model for weighting discipline and learning outcomes (Mean Squared Error [MSE]: 2.9224, Root Mean Squared Error [RMSE]: 1.4252, Coefficient of Determination [R2]: 0.9667), outperforming three alternatives. Gradient Boosting Machine (GBM) was best for selecting three competencies (MSE: 0.0221, RMSE: 0.1093, R2: 0.9997). These models outperformed methods such as Principal Component Analysis (PCA), K-Means, Random Forest (RF), Matrix Factorization (MF), K-Nearest Neighbors (KNN), and Neural Collaborative Filtering (NCF). The integration of GB and GBM produced the CR model, validated using internal and external datasets, showing consistent performance. The findings underscore the role of attendance in shaping personalized, competency-driven pathways. This study advances AI-driven, competency-based education (CBE) by integrating behavioral and academic metrics into a scalable, data-informed recommendation framework.