Feature Identification from Imbalanced Data Sets for Diagnosis of Cardiac Arrhythmia

Considering the problem of class imbalance, the method of feature identification for disease diagnosis based on equalization methods and elastic net is proposed. First, the cardiac arrhythmia dataset from UCI machine learning repository is dealt with by random under-sampling method and synthetic minority over-sampling technique (SMOTE). Then, the feature sequence of the dataset is identified by elastic net. Finally, “one-against-one” classification strategy of multi-classes SVM classifier is employed to test the classification accuracies of different diagnosis rules. Experiments show that the average accuracy achieves 77.36%, when the first nine attributes in the feature sequence are used as diagnostic rules. Compared with other approaches in the literatures, the presented algorithm is more effective in removing redundant features and can better identify the type of arrhythmia.