A Novel Feature Selection Algorithm for the Detection of Obstructive Sleep Apnea by Using Heart Rate Variability and ECG Derived Respiratory Analysis

Obstructive Sleep Apnea (OSA) is a commonly known sleeping disorder whose undiagnosed and untreated condition can be fatal to cause cardiac fibrillation, arrhythmia and stroke. The recent study aims for a novel computer based methodology for the automated detection of OSA by considering two biosignals from electrocardiogram (ECG) such as Heart Rate Variability (HRV) and ECG-derived Respiratory (EDR). The input signal is retrieved from a publicly available database by using Polysomnography (PSG). Then, the collected data is filtered and pre-process to acquire HRV and EDR parameters from the ECG channel. Three supervised feature selection algorithms, namely Pearson’s Correlation Coefficient (FC), ReliefF and Mutual Information Gain Maximization (MIGM) to obtain the optimal features and to balance the feature dimensionality. The input signals are fed into an ensemble learning algorithm for evaluating the effectiveness of proposed feature selection algorithms. The results are compared on the performance matrices and it is observed that MIGM algorithm provides the most optimal features compared two other two selection techniques. Also, the model is evaluated for temporal features and frequency domain features derived from HRV and EDR signals accordingly. The feature domains are also compared depending on their model performance. The comparison states that temporal features are better in accuracy as compared to frequency domain features, however, combination of both the features have improved the validation significantly. Therefore, our proposed methodology acts in a promising way to use the feature selection algorithms and multi-modal analysis for the accurate detection of OSA.