Dimensional Reduction to Improve Learning Recommender Systems Precision Based on Grades

The incorporation of optional subjects into higher education curricula is a ubiquitous practice, affording students the latitude to pursue coursework aligned with their interests across diverse knowledge domains such as humanities, arts, sciences, and technology, along with supplementary offerings. Despite this expansive selection, students often grapple with an inundation of information when confronted with selecting optional subjects. Recognizing the extensive dataset encompassing student grades and subject enrollment, this study proposes the implementation of a Learning Recommender System, based on collaborative filtering, a machine-learning technique. This method harnesses students’ academic success in particular subjects to proffer judicious recommendations to their counterparts. A challenge encountered by recommender systems is the issue of data sparsity, given that not all students undertake every available course, resulting in a sparse matrix. To surmount this challenge, the study advocates for the application of dimensional reduction techniques, which serve to transform the sparse matrix into a more condensed yet informative representation. This reduction augments the predictive efficacy, and versatility for employing a spectrum of machine-learning techniques. The Matrix Factorization model emerges as the preeminent choice in contemporary recommender systems. The research substantiates the efficacy of these methodologies through an empirical examination of data sourced from Salesian Technical University, comprising records of ten thousand students and one thousand subjects over a five-year duration. In essence, the study conclusively demonstrates that dimensional reduction is indispensable for facilitating the data sparsity predicament in recommender systems, thereby enhancing the efficiency and efficacy of the proposed Learning Recommender System.