hybridization of Random Forest & Deep learning for prediction of maternal mortality with socio economic data in katsina state

Abstract

This study presents a hybrid machine learning model that integrates Random Forest Regression with an Artificial Neural Network (ANN) to predict maternal mortality using socio-economic data. The model leverages key features such as maternal education, maternal age, antenatal visits, and an interaction term between education and antenatal care (ME × AC). The random forest, configured with 200 estimators, first generates predictive insights, which are then appended to the original features to form an enriched hybrid input space for the ANN.

Standardized features and an adaptive learning rate are used to train the ANN, which is made up of three hidden layers (128–128–64 neurons) and ReLU activations. Strong regression performance is demonstrated by the hybrid model's Mean Squared Error (MSE) of : 0.0748 and R-squared (R²) value of 0.6663. With regard to classification (threshold = 0.41), it records 90% accuracy, 89% precision, 96% recall, 92% F1-score, and 0.95 area under the ROC curve (AUC). These findings show that the model can produce precise, comprehensible estimates of maternal mortality based on socioeconomic factors, providing a useful instrument for public health planning in settings with limited resources.