A Hybrid Attention-based LSTM-XGBoost Model for Detection of ECG-based Atrial Fibrillation

Abstract

Atrial fibrillation (AF) is a frequently encountered heart arrhythmia problem today. In the method followed in the detection of AF, the recording of the Electrocardiogram (ECG) signal for a long time (1-2 days) taken from people who are thought to be sick is analyzed by the clinician. However, this process is not an effective method for clinicians to make decisions. In this article, various artificial intelligence methods are tested for AF detection on long recorded ECG data. Since the ECG data is a time series, a hybrid model has been tried to be created with the Long Short Term Memory (LSTM) algorithm, which gives high results in time series classification and regression, and a hybrid method has been developed with the Extreme Gradient Boosting algorithm, which is derived from the Gradient Boosting algorithm. To improve the accuracy of the LSTM architecture, the architecture has been strengthened with an Attention-based block. To control the performance of the developed hybrid Attention-based LSTM-XGBoost algorithm, a public data set was used. Some preprocessing (filter, feature extraction) has been applied to this data set used. With the removal of these features, the accuracy rate has increased considerably. It has been proven to be a consistent study that can be used as a support system in decision-making by clinicians with an accuracy rate of 98.94%. It also provides a solution to the problem of long ECG record review by facilitating data tracking.