Unbiased Mortality Prediction for Unbalanced Data Using Machine Learning

Mortality defines `to be in the state of being prone to death'. The mortality rate is the measure of the frequency of occurrence of death in a defined population during a specified interval. The major causes of death in the recent decade is due to the Cardiovascular diseases, Cancers and Respiratory diseases. The problem is hosted on CodaLab as an ongoing competition. In this work, we have been provided with a data set that contains 342 features. The training data consists of 80,000 patient's records which are highly biased as 90 percent of the labels belong to the class `0'. Due to its skewed nature, various oversampling techniques were tried. Certain Machine learning techniques were then applied to detect the death chances of a person during their stay at the hospital based on their medical records. Accuracy of 76.68% percent was achieved on the testing data (20,000 records).