An Improving ICU Patient Mortality Prediction Using Artificial Neural Networks
Nowadays Physicians facing a lot of issues while treating ICU patients. Recording the severity of illness of patients who are in ICU can give an evaluation of the patient’s condition and thus help doctors to serve what kind of treatment and medicines. This study aims to develop an artificial neural network model for predicting patient mortality in hospitals. Dataset contains the features which are used to predict the mortality of ICU patient. It consists of 4000records. Dataset is divided into train dataset and test datasets Twenty-six features are taken after preprocessing techniques. The correlation heat map is used to find the relation among attributes. The training set is provided with outcomes. Using certain ICU patient parameters, a neural network model is developed to predict the mortality of ICU Patients in the hospital. A sequential model of Keras is used to develop a neural network model. The dense layer and drop out layer are used to develop a neural network model. Rectified linear unit and sigmoid activations are used. ModelCheckpoint and ReduceLROnPlateau callbacks are used to build the model without overfitting. Batch Normalization is used to normalize the dataset. RMSProp and Adam Optimizer are used to optimize the output. The accuracy of predicting mortality of ICU patients is about 89%.