ECG Classification with Beat-By-Beat Fusion Features for Cardiac Disease and Apnea Detection in Smart Health

Cardiac disease is a condition that affects the heart and can be prevented or treated through a healthy lifestyle. Treating apnea can improve heart failure-related conditions. We combined MITBIH arrythmia database and Apnea-ECG database to improve ECG classification performance for detecting various cardiac diseases and apnea. We created three feature modules and applied principal component analysis (PCA) and analysis of variation (ANOVA) feature selection algorithms to identify top ranked features, separately. Top-ranked features are added sequentially and compared with all features for accuracy. We explored various machine learning and deep learning models, including KNN, SVM, Naive Bayes, Decision Trees, Bagged Trees, ANN, RNN, LSTM, and CNN to evaluate their performance. Features selected by ANOVA had higher accuracy than features selected by PCA. The accuracy of 96% was achieved using the first 11 features selected by ANOVA in Module 2 and the accuracy of 98% was achieved by combining the features selected using PCA and ANOVA in Module 3. Our results showed that module 3 which is combination of signal independent feature and specific features improves the accuracy and effectiveness of ECG classification for smart heart monitoring wearable devices. Our beat-to-beat classification method reduces the number of features used in real-time classification of wearable ECG data.