Comprehensive Performance Assessment of Multi-Neural Ensemble Model for Mortality Prediction in ICU

The development of models to estimate the mortality rate of critically ill patients in the intensive care unit(ICU) is significantly enhanced by technologies based on artificial intelligence. But it can be quite difficult because most medical data are typically unbalanced and incomplete. This paper aims to address the missing data using advanced imputation algorithm like single center imputation from multiple chained equation (SICE) and synthetic minority oversampling technique with edited nearest neighbor (SMOTE-ENN) method is used to solve the imbalanced data problem. Wrapper-based genetic feature selection method is used for the feature selection. With a highly skewed dataset, the objective of this work is to propose an improved ensemble classifier that utilizes a combination of boosting and stacking ensemble strategies. Our work proposed a Stacking-based multi-layer perceptron ensemble learning method on bench-marking dataset namely 2020 WiDS (Women in Data Science) Challenge and got an accuracy of 93.67%. The results obtained using randomized five-fold cross-validation and hold-out techniques reveal that the performance of the proposed ensemble algorithm was better than that of all other predictors. In comparison to prior deep learning-driven mortality classification research, we designed a comprehensive structure that encompasses a novel feature pre-processing methods and stacking ensemble algorithm for classification.