An ANN-Based Predictive Model for Early Detection of Maternal Health Risks

Abstract

Maternal health complications are also a major issue in many countries of the world especially those with very few resources since in most cases early detection of high-risk pregnancy is not easy. This paper introduces a predictive model based on Artificial Neural Network (ANN) to predict maternal health risks in early stages using the publicly available Maternal Health Risk Dataset. The model uses clinical and physiological characteristics such as age, blood pressure, blood sugar, body temperature, heart rate and trimester of pregnancy. Preprocessing of data was done to address missing values, feature scaling, categorical encoding, and class balancing with the Synthetic Minority Oversampling Technique (SMOTE). The second model was a feedforward multi-layer ANN where the activation was ReLU, dropout regularisation, and softmax output implemented in TensorFlow and Keras. The model was trained on Adam optimizer and tested on the basis of accuracy, precision, recall, F1-score, and confusion matrix. Results show the ANN achieved 94.2% training accuracy, 92.8% validation accuracy, and 93.6% accuracy on the test set, with a high recall of 0.94 for the high-risk class, indicating reliable identification of pregnancies at greatest risk. These results indicate that ANN-based predictive modelling has the potential to improve maternal health surveillance, help promote clinical intervention, and minimise preventable maternal morbidity and mortality.