Hybrid Deep Learning Model for Predicting Mortality Due to Diarrheal Diseases in Children

Abstract

Diarrheal diseases remain a leading cause of mortality among children under the age of five, particularly in low-resource regions. Early prediction and intervention are critical for reducing these preventable deaths. We apply advanced deep learning models (standalone such as Convolutional Neural Networks (CNNs), Multilayer Perceptron’s (MLPs), Autoencoders as well as hybrid models CNN + MLP and Autoencoder + CNN) to predict high and low death rates of the diarrheal diseases. The Preprocessing of the dataset was done and then converted into a binary classification task given from global mortality data. Results demonstrate that hybrid models outperformed standalone approaches, achieving validation accuracies exceeding 99%, with consistent precision, recall, and F1-scores for both high and low death rate predictions. The learning curves indicate robust training processes with minimal overfitting, supported by regularization techniques such as dropout. This study highlights the potential of hybrid deep learning architectures in capturing both local and global patterns within mortality datasets, providing an effective tool for early prediction and intervention in public health initiatives aimed at reducing child mortality.