Predicting Maternal Health Outcomes Using Decision Tree Models: Results and Limitations from Rural Sudan

Abstract

This study addresses the critical challenge of maternal health resource misallocation in rural Sudan, which significantly impacts maternal mortality rates. Leveraging data mining techniques, specifically the Decision Tree algorithm, pregnancy care data (n=2,201) collected by the Ministry of Cabinet Central Bureau of Statistics were analysed. The primary objective was to develop a predictive model to classify the outcomes of the last completed pregnancy (Alive, Dead, or Abortion) to support evidence-based decision-making. Utilising a hold-out test set of 331 instances, the model achieved a 100% classification accuracy, perfectly identifying outcomes for 293 ‘Alive’, 30 ‘Dead’, and 8 ‘Abortion’ cases. The findings highlight key predictors, including geographic location, access to professional antenatal care (ANC), and the importance of routine diagnostic tests such as blood sampling. The study provides actionable insights for policymakers to optimise resource allocation and enhance maternal survival through targeted interventions in high-risk rural areas.