Application of Artificial Neural Networks and Conventional Statistical Methods in Predicting Maternal Mortality in Maiduguri, Nigeria

In Maiduguri, a conflict affected area in North East Nigeria, maternal mortality has been a serious public health issue in the region. This study examined socio demographic and obstetric factors associated with maternal mortality and examined the usefulness of predictive modeling based on collected maternal health data. A facility based observational study was conducted among 158 women who accessed maternal health services. Descriptive statistics and chi-square tests were used to assess relationships between maternal mortality and selected variables, including age, marital status, educational level, and antenatal care attendance. Maternal mortality was observed in 25.3% of cases. Maternal age, educational attainment, and number of antenatal care visits showed statistically significant relationship with maternal mortality (p < 0.05), whereas no statistically significant relationship was observed with marital status. Most maternal deaths occurred among women with no formal education and those who attended fewer antenatal visits. A multilayer perceptron model demonstrated good classification performance, with an accuracy of 98.1% in the training sample and 88.2% in the testing sample, and a high ability to identify maternal deaths. These findings indicate a persistently high burden of maternal mortality in Maiduguri and emphasize the importance of improving antenatal care utilization and female education, while suggesting that predictive approaches may complement conventional methods for identifying women at increased risk in resource-limited settings.