ECG Measurement System for Vehicle Implementation and Heart Disease Classification Using Machine Learning

Abstract

The 12-lead electrocardiogram (ECG) method can diagnose more cardiovascular disease than the single-lead method, but it is difficult to use in daily life because numerous electrodes must be attached to the body. As an aging society approaches, electrocardiography is expanding its use in daily life, even to the heart disease monitoring system for passengers in vehicles. This study proposes a single-lead ECG measurement system implemented in the steering wheel of the vehicle and a machine learning model to classify the driver’s heart health status. For the ECG measurement system, an algorithm to obtain stable ECG signals is proposed along with the measurement hardware. It uses the range and interval of the ECG signal to determine stability under noisy conditions caused by vehicle vibration and the driver movement. To classify four classes of heart diseases (normal, atrial fibrillation, other rhythms, noise), a two-stage machine learning structure is proposed. To train the machine learning models with an optimal feature subset, 188 features were extracted from the single-lead ECG dataset, and a sequential wrapper-type feature selection was conducted. As a result, when the naïve Bayes model using ten features was located in the first step and the support vector machine using 13 features in the second step, the proposed two-stage classification structure returned the best score ${F1}_{NAO}=0.7898$ and real-time classification performances (0.86 seconds on average).

Keywords

• Atrial fibrillation
• ECG
• feature selection
• electrocardiogram
• health care
• machine learning