Optimising the prediction of malnutrition risk using semi-supervised learning: a case study of PCIMA data from Eastern Kasai (DRC)

Abstract

Malnutrition remains a major public health challenge in the Democratic Republic of the Congo (DRC), particularly in the province of Kasaï Oriental, where rates of acute and chronic malnutrition remain high. However, the limited availability of properly labelled data constitutes a significant obstacle to the development of high-performance predictive models to aid decision-making in the planning of nutritional interventions.

This study proposes a semi-supervised learning approach combining an unsupervised clustering algorithm (K-Means) and two supervised classifiers (Support Vector Machine and K-Nearest Neighbours) to improve the prediction of malnutrition risk. The data used were drawn from the Integrated Management of Acute Malnutrition (IMAM) programme, collected across the 19 health zones of Kasaï Oriental between 2020 and 2022.

The methodology comprises four stages: (i) data pre-processing and normalisation, (ii) clustering of observations using K-Means to generate pseudo-labels, (iii) supervised learning using SVM and KNN models, and (iv) weighted fusion of the predictions produced by the various algorithms. The model’s performance was evaluated using accuracy, precision, recall and the F1 score.

The results show that the proposed hybrid approach significantly improves predictive performance compared with individual models. Projections for the period 2023–2026 identify several health zones at high risk of a deterioration in the nutritional situation. This research thus demonstrates the relevance of semi-supervised learning for predictive modelling in contexts characterised by a scarcity of labelled data.

Keywords: Semi-supervised learning, Malnutrition, Machine learning, K-Means, SVM, KNN, PCIMA, DRC.

Received Date: August 17, 2026

Accepted Date: September 09, 2026

Published Date: October 01, 2026

Available Online at: https://www.ijsrisjournal.com/index.php/ojsfiles/article/view/962