Predicting Middle School Students' Academic Orientation Using SOM and Machine Learning
Predicting Middle School Students' Academic Orientation Using SOM and Machine Learning

Summary/Abstract: Background/purpose. Middle school is a critical stage for shaping students' academic paths, but traditional orientation methods often fail to predict suitable trajectories, leading to mismatches that impede success. This study aims to develop a proactive, data-driven framework to forecast academic orientation for middle school students, enhancing tailored educational guidance. Materials/methods. The study utilized Self-Organizing Maps (SOM) and random forest prediction to analyze data from 720 Moroccan middle school students. In Phase One, survey responses (e.g., interest, selfefficacy) and math/science scores were clustered using a 7x7 SOM grid. In Phase Two, a random forest classifier (150 trees, max depth = 12) was trained on 70% of the data (504 students) with 17 features to predict orientation outcomes, validated with statistical tests (ANOVA, chi-square). Results. SOM identified five profiles: Cluster 1 had high math scores (Mean = 16.5) and 85% STEM preference; Cluster 3 had lower scores (Mean = 9.5) and 75% literary inclination with anxiety. Random forest achieved 93% training (F1 = 0.92), 89% test (AUC = 0.94), and 87% validated accuracy, predicting 57% scientific and 43% literary tracks. Self-efficacy and math scores predicted scientific paths; anxiety drove literary choices. Conclusion. This framework outperforms traditional methods, enabling early, personalized orientation. Despite some misclassification, counselor feedback (80% agreement) supports its utility. Future refinements could enhance accuracy and equity in student outcomes.

• Details
• Contents