Authors: Brian Levi Okimaru, Betty Mayeku, Humphrey Juma kilwake
Abstract: The transition from junior to senior school under Kenya's Competency-Based Education (CBE) requires learners to select academic pathways that align with their competencies and interests. This transition presents a challenge because pathway selection requires personalized guidance while ensuring the privacy of sensitive student information. Existing educational recommender systems predominantly rely on centralized data processing, exposing learner data to privacy risks and limiting the secure exchange of information across institutions. This study proposes a privacy-preserving personalized pathway recommender system that integrates federated learning, cosine similarity, and Random Forest to support academic pathway recommendation without sharing raw student data. Cosine similarity was employed to model learner competency profiles and measure their alignment with predefined pathway requirements. The resulting similarity scores were incorporated into a Random Forest classifier through feature engineering to improve pathway prediction accuracy. A horizontal federated learning framework enabled multiple schools to collaboratively train the recommendation model by exchanging only model updates while retaining student records locally. The proposed model was evaluated using accuracy, precision, recall, and F1-score. Experimental results showed that integrating cosine similarity with Random Forest improved pathway classification performance, while the federated recommender system achieved an accuracy of 86.54%, outperforming the centralized recommender approach while preserving student privacy. The proposed framework provides an effective and privacy-preserving decision-support tool for personalized academic pathway recommendation within Kenya's Competency-Based Education. The study demonstrates that integrating federated learning with content-based filtering and machine learning can simultaneously enhance recommendation accuracy, personalization, and data privacy in educational environments.
DOI: http://doi.org/10.5281/zenodo.21126397
