Article Details

Teaching Practice (TP) assessment is a fundamental component of teacher education in Nigeria because it determines the professional readiness of pre-service teachers. Despite the use of standardised assessment rubrics, TP evaluation remains vulnerable to subjectivity and potential inconsistencies in scoring among supervisors. This study developed a predictive modelling framework for detecting anomalies in TP scoring using machine learning techniques. Secondary TP assessment data comprising 2,685 records from 102 supervisors, covering seven weighted assessment components: Preparation, Presentation, Classroom Management, Communication Skills, Evaluation, Teacher Personality, and Reflective Journal, were analysed using the Extreme Gradient Boosting (XGBoost) regression algorithm, with an 80:20 train–test split. The model was used not to predict scores for their own sake, since the total score is arithmetically derivable from its components, but to learn a behavioural benchmark of typical supervisor scoring against which each supervisor’s actual awarded scores could be compared through residual analysis. On the held-out test set, the model achieved a Mean Absolute Error (MAE) of 2.13 and an R² of 0.89. Residual analysis identified a small subset of supervisors whose scoring patterns persistently diverged from the learned benchmark, suggesting possible scoring inconsistency that warrants moderation review rather than a confirmed bias. Feature importance analysis showed that Presentation and Teacher Personality contributed most strongly to predicted scores. The study contributes a supervisor-level moderation framework, novel in Nigerian teacher education, that supports evidence-based, targeted review of TP assessment records rather than blanket manual rechecking, and offers a scalable approach to improving assessment reliability in Nigerian Colleges of Education