Experimental comparison of photoplethysmography-based atrial fibrillation detection using simple machine learning methods

The results of experimental studies on application of selected simple machine learning (ML) methods for detection of atrial fibrillation (AFib) based on photoplethysmogram (PPG) are presented in the paper. The goal of the studies was to compare the performance of AFib detection using different ML algorithms in short PPG segments containing 32 consecutive cardiac cycles. Four parameters describing time series of interbeat intervals (IBI) were derived from the time domain Heart Rate Variability (HRV) and used as features for classification algorithms. Optimal values of metaparameters for all considered ML algorithms were experimentally determined. Accuracy, sensitivity, specificity and F1-score were then calculated to measure the quality of detection performance of each classification algorithm.