Machine Learning Models for the Detection of ADHD in Primary School Students in Peru

Vara J.; Dioses D.; Wong L. · Conference of Open Innovation Association, FRUCT (2025) · DOI: 10.23919/fruct67853.2025.11239195

Resumen

Attention-deficit/hyperactivity disorder affects a notable share of the school-age population, and its timely identification remains challenging in educational settings with limited resources. This work presents a supervised learning approach to support early screening, comparing three classification models: a tree ensemble, a margin-based classifier, and a multilayer neural model. The process follows a standard data mining methodology, covering problem understanding, data understanding, preparation, and modeling. The dataset comprises 139 digital questionnaires completed by parents, guardians, and teachers, focusing on behavioral and contextual indicators of primary school students with and without a clinical diagnosis. The three models showed competitive performance. The tree ensemble offered a balanced trade-off between positive-class detection and overall correctness and achieved the lowest logarithmic loss, indicating more reliable probability estimates. The neural model reduced missed positive cases, which is desirable when minimizing false negatives in school contexts. These results suggest that a questionnaire-based, low-cost pipeline can provide practical decision support for early identification in resource-constrained educational environments.

Instituciones: Universidad Peruana de Ciencias Aplicadas