Modelling the Determinants of Maternal Mortality: A Comparative Study of Logistic Regression and Artificial Neural Network Model

Abstract

Without any doubt, Indicators derived from mortality rates give a clear representation of the overall population health. One of such indicators is maternal mortality. This work discusses Logistic regression and Artificial Neural Network model and the application of these models in predicting maternal mortality. 276 records (ranging from 2003 to 2012) on mother's age, mode of delivery, parity, sex of the baby, baby's weight at birth, nature of complication (independent variables) and mother's status (dependent variable) were collected from the medical record department of the University College Hospital Ibadan. Logistic regression model was used to check for the risk factor associated with maternal mortality. In order to compare the efficiency of ANN and logistic regression model, the following measures were used: sensitivity, specificity and goodness of fit. Results of the analysis revealed that parity and age are the major determinant of maternal mortality. The result of the comparison of the efficiency of ANN and logistic regression model showed that ANN outperformed logistic regression with sensitivity 50.6% versus 31.0%, specificity 91.6% versus 86.6% and the mean square error (MSE) of ANN is very small compared to that of the logistic regression model.