Exploring structural determinants of mathematics performance in Tanzanian secondary schools using machine learning approach

Mathematics performance in Tanzanian secondary schools remains critically low, with form four national examination pass rates averaging 20.87% between 2018 and 2023. Majority of existing studies attribute such poor performance to resource deficits, psychological, social, economic, and instructional dimensions. However, up to now a few studies have applied machine learning techniques to rigorously identify the structural determinants of mathematics performance in Tanzanian secondary schools. This study employed Lasso Regression and Principal Component Analysis (PCA) on data collected from three thousand and two (3,002) form three students across five geographical zones in Tanzania. From 88 initial variables, Lasso Regression identified 35 significant predictors grouped into four categories: psychological factors, instructional quality, household assets, and demographic and environmental factors. The results reveal that Math Career Utility and Math Confidence are the strongest positive predictors, while Smartphone ownership, Commute Time, and Chore Hours significantly hinder performance. PCA identified Teacher Clarity, Adaptability, and Encouragement as a unified instructional construct. These findings challenge resource-centred assumptions and offer evidence-based recommendations for policy intervention in mathematics education in Tanzania and across sub-Saharan Africa.