AutoML Dropout-Risk Model for Tanzania's Secondary Schools — University of Dar es Salaam

Tanzania's secondary schools have struggled with a persistent and slightly worsening dropout problem, with the national rate rising from 3.8% in 2018 to 4.2% in 2019, and with limited formal tools available to identify the root causes or project which students are most at risk.

Researchers Yuda N. Mnyawami, Hellen H. Maziku and Joseph C. Mushi, in the Department of Computer Science and Engineering at the University of Dar es Salaam, responded with an Automated Machine Learning (AutoML) approach that selects the hyperparameters, features and algorithm best suited to the available data, rather than relying on a single hand-tuned model. They trained and tested the approach on Tanzania's Twaweza Uwezo household learning-assessment dataset, which covers households across the country, comparing four classifiers.

The reported accuracy was high across all four models — Decision Tree 99.8%, K-Nearest Neighbours 99.6%, Multi-Layer Perceptron 99% and Naive Bayes 97% — with the authors arguing that automated hyperparameter and feature selection outperformed the manually tuned models used in earlier dropout-prediction studies.

These figures should be read cautiously: accuracy this high is often a sign of a small or imbalanced evaluation set rather than a production-ready classifier, and the paper does not report validation against an independent Ministry of Education dataset or a real classroom pilot. A related follow-up by overlapping authors extends a similar Bayesian-optimisation approach to secondary schools elsewhere in Sub-Saharan Africa, suggesting this is part of an active regional research programme — but as of publication, no ministry has deployed the model as an operational early-warning tool, and no intervention outcomes have been measured.