A Machine Learning Approach to Detect Changes in Single-Lead Electrocardiogram Signals Post Coffee Consumption

Abstract

Electrocardiograms (ECGs) are the most reliable method for detecting changes in heart function. It reflects the heart's electrical activities and can be interpreted using various waves, peaks, and intervals. Several factors influence the heart's functionality, including lifestyle, stress, food, etc. The world's most extensively consumed beverage, coffee, is a vital component of daily living. Caffeine, the primary component of coffee, is believed to influence the physiology of the heart. However, the influence of caffeinated coffee consumption on cardiac electrophysiology, as assessed from morphological features (e.g., peaks, waves, intervals), is ambiguous, as the results are inconsistent. This has led to the investigation of alternative feature extraction approaches for properly detecting changes. The current study focuses on evaluating single-lead electrocardiogram (ECG) data to determine if there are any significant changes in the ECG signal following coffee drinking and then on predicting these changes using various machine learning models. The ECG signals of human participants were obtained both before and after the consumption of coffee. 1260 ECG segments were extracted from the two data groups (control and stimulus) for further investigation. A preliminary investigation on the ability of statistical and entropy characteristics to detect coffee-induced changes was first performed. The statistical analysis yielded favorable results. These sets of features were then extracted from the ECG signals after employing three distinct decomposition techniques: discrete wavelet transform (DWT), wavelet packet decomposition (WPD), and continuous wavelet transform (CWT). Multiple machine learning (ML) models used these extracted features as input, and the classification performance for predicting the coffee-induced alterations in the ECG signal was examined. After utilizing the decomposition method, the classification accuracy appears to have improved. In addition, the CWT approach was found to be more accurate than the other two decomposition methods in predicting the onset of any change in the ECG signal after coffee consumption. In the future, the current work may potentially prove helpful for identifying changes in cardiac activity following consumption of other caffeinated beverages (e.g., tea, cola, soft drinks, etc.), medications, and alcohol.