Multi-Class Electrogastrogram (EGG) Signal Classification Using Machine Learning Algorithms

Electrogastrogram (EGG) is a simple and non-invasive method in clinical practices for assessing the stomach function by observing the gastric myoelectrical activity extracted using the electrodes placed on the abdominal surface. EGG is a slow wave propagation. Based on the dominant frequency or cycle per minute, there are three types of EGG signals: Normogastria, Bradygastria, and Tachygastria. In this study, we used the Logistic Regression (LG), Support Vector Machine (SVM) and K Nearest Neighbor (KNN) Machine Learning (ML) algorithms to successfully classify two and three types (classes in ML terminology) of EGG signal with high accuracy. Our results show that the SVM algorithm performs best to classify the two and three class signals with an accuracy of 100% and 92.11% respectively, while logistic regression and the KNN algorithms demonstrate similar lower performances. SVM algorithm also achieved a maximum F1 score, precision, and recall value of 100% and 92% for the two and three classes of EGG signal respectively. An Area Under the Curve (AUC) score of 100% and 92% are observed in the two-class and three-class problem respectively in EGG signal classification using the SVM algorithm. Based on our analysis, we can conclude that SVM can be implemented successfully to accurately classify multi-class EGG signals.