Cardiac Arrhythmia Identification Using Feature Selection and Rule-Based Classifiers

Nowadays, one of the main causes of death is cardiac arrhythmia. The computational identification of cardiac arrhythmia from ECG signal recordings is vital for timely clinical diagnosis and treatment. This article presents an approach to evaluate the performance of various feature selection methods using different machine learning classification techniques to detect cardiac arrhythmia. We used ECG recordings from a data set with 279 features from the UCI repository. To generate optimal performance, we use the relevant variables identified through the CFS, Consistency, Chi-Square, Information Gain, Symmetric Uncertainty, Gain Ratio, and OneR filter methods. These variables were then the input for the C4.5, Rpart, PART, JRip, and OneR classifiers. In the experiments, we found a set of 37 relevant features with an average balanced accuracy of 77.23% using C4.5 and CFS, without applying any data balancing method.