Heart Failure prediction using Xgboost algorithm and feature selection using feature permutation

Heart is a vital organ of the human body which provides body with oxygen rich blood. If the heart stops working, then a human being dies within a span of few minutes. This paper analyses the performance of Xgboost algorithm in mortality prediction among the heart failure patients. The performance of Xgboost is compared with various other prediction algorithms. Hyperparameter tuning has been performed on the Xgboost algorithm to improve the model performance and to reduce overfitting. In the second part of this work, feature importance using feature permutation have been performed using which important features from the dataset has been selected . In hyperparameter tuning of the algorithm sklearn’s RandomizedSearchCV have been used which randomizes the search on the hyperparameters. Stratified folds is used during hyperparameter tuning to make sure that training and test data remains well balanced during model creation and prediction. Using feature selection, it has been established that using a smaller number of important features can improve the model performance by significant amount and can reduce the overfitting. In this work accuracy has been used as the performance metrics.