Automated Congestive Heart Failure Detection Using XGBoost on Short-term Heart Rate Variability

Congestive heart failure (CHF) is one form of cardiovascular diseases (CVD) which have caused an estimated of 17.9 million deaths in 2021. Heart rate variability (HRV) is found to be correlated with various heart diseases. With the advance of artificial intelligence, many have tried to develop an automated CHF detection model using machine learning algorithm and HRV features. In this paper, we built an automated CHF detection using XGBoost model to distinguish between CHF and normal subjects using short-term HRV. Our model was trained on three different combinations of public datasets: 1) CHFDB-CHF2DB-NSRDB-NSR2DB, 2) CHFDB-NSRDB, and 3) CHF2DB-NSR2DB. Grid Search method with Stratified K-Fold Cross Validation was used to find the best set of hyperparameters. The grid search was run with three different scoring objectives to maximize: accuracy, sensitivity, and specificity. Across the three different dataset combinations, the model trained to optimize sensitivity achieved the overall best results in all of them. For the first combination, the model achieved 0.966 accuracy, 0.977 sensitivity, and 0.963 specificity on the test set. For the second combination, the model achieved 0.920 accuracy, 0.797 sensitivity, and 1.0 specificity. And for the third combination, the model achieved 0.932 accuracy, 0.870 sensitivity, and 0.939 specificity. Our analysis on the feature importance of the best models of trained from each combination also showed that Age, LF/HF, and NNI_20 are consistently ranked high by the models to distinguish the existence of CHF.