Benchmarking the Neuro-Boosting Fusion Network (NBFN) with Existing Models for Predicting Maternal Mortality and Morbidity Risks

The globe has witnessed large scale problems related to pregnancy. In particularly, there have been numerous cases of death caused by various forms of severe obstetric complications. Given these challenges, a model capable of making an accurate and professional interpretation regarding maternal risk evaluation will be very useful towards ensuring adequate intervention in time. Various types of Machine Learning and Deep Learning models can provide good contributions towards solving some of the global problems affecting mother?s health. For instance, models based on logistic regressions and Decision Trees have provided significant assistance towards achieving good maternal health outcomes. Nevertheless, most ML and DL models use isolated structured variables or face limitations around interpretability. A solution to this limitation was presented through the proposal of a Neuro-Boosting Fusion Network (NBFN). This network integrates a Keras powered Convolutional Neural Network (CNN) designed to extract features automatically from image input, combined with an Extreme Gradient Boosting (XGBoost) Classifier for carrying out the final stage of risk predictions. This paper evaluated the performance of our NBFN model against several other benchmark and state-of-the art models. Results achieved via evaluating all risk assessment models are reported in Chapter eight using multiple parameters namely Accuracy, Precision, Recall, F1 Score, AUC-ROC (Area Under the Receiver Operating Characteristic Curve), Runtime, Interpretability amongst others with their respective statistical analysis conducted using 5-Fold Cross-Validation. According to the results obtained during the experiments performed on the dataset provided by the hospital, the model demonstrated superior accuracy scores at 95.80% and 94.90% recall rate. Also, the Area Under the ROC curve stood at excessive rating of 0.97 proving each the medical efficacy of the framework collectively with the computational robustness of the mannequin appropriate adequate for deployment as a choice assist device for maternal care.