Circulatory Failure Prediction by Using Wavelet Transform and Machine Learning Methods

In order to predict the circulatory failure ahead, a method is proposed by combing wavelet transform and machine learning methods AdaBoost, XGBoost, KNN and SVM. Wavelet transform can decompose the original time series into smooth and detail information on different resolution levels, which can detect the autoregression effect involved in the time series that can not be detected by these methods directly. By applying the method to a ICU dataset which contains 1718 patients and 18 variables, the results show that the methods AdaBoost and XGBoost gain good performance with ROC score around 0.96 when the time series is predicted ahead by 5 minutes, 25 minutes or 50 minutes. For patients under different basic clinical information, results show that age play an important role in model performance improvement, followed by sex, weight and height.