<![CDATA[Mathematics achievement in Tanzanian secondary schools remains persistently low, yet national examination reporting is largely descriptive and offers limited insight into how performance is structured across schools. This study integrated classical statistical inference with machine learning to examine school-level mathematics performance using 2024 Certificate of Secondary Education Examination (CSEE) data published by the National Examinations Council of Tanzania (NECTA). The analytic sample comprised school-level grade distributions (Grades A, B, C, D, and F) and candidate totals for 22 schools drawn from four regions: Dar es Salaam, Morogoro, Mwanza, and Mtwara. Analyses were conducted in MATLAB R2022b and combined one-way analysis of variance (ANOVA), k-means clustering, a pruned regression decision tree, Pearson correlation analysis, principal component analysis (PCA), and multiple linear regression. The ANOVA yielded no statistically significant difference in overall mathematics performance across regions, F(3, 20) = 2.75, p = 0.07, a near-threshold result that should be read cautiously given the limited number of schools. Clustering nevertheless partitioned schools into three groups of markedly different grade profiles, indicating that variation within regions was more pronounced than variation between them. The regression tree identified the number of candidates attaining Grade A as the primary splitting variable, with Grades C and D contributing secondary distinctions. Correlation analysis showed that adjacent grade categories co-varied positively while top and bottom grades were inversely associated, and PCA indicated that no single component dominated the grade structure. A multiple regression model explained 90.4% of the variance in the composite performance score, with Grade A and Grade C counts emerging as the strongest predictors; because total enrolment is the exact sum of the grade counts, the model was rank-deficient and its explanatory power is partly definitional. The findings suggest that raising school mathematics performance requires simultaneous attention to the upper tail and to the large body of moderately achieving learners, and they demonstrate both the promise and the interpretive limits of applying integrated statistical and machine learning methods to aggregated national examination data.]]>
Copyrights © 2026
