Gender Classification Using nonstandard ECG Signals - A Conceptual Framework of Implementation

This paper presents a comparison of nine models for gender identification using nonstandard ECG signal. Methods: QRS features, QT interval, RR interval, HRV features and HR were extracted from three minutes of 40 ECG’s (from 24 female and 16 males) available at ALLab dataset and 108 ECG’s (from 52 female and 56 males) available at CYBHi dataset. Models were developed using Decision tree, SVM, kNN, Boosted tree, Bagged tree, Subspace kNN, Subspace Discriminant, two majority vote and verified by external validation. Results: The study presented achieved as best results an accuracy of 78% from Boosted tree and 85% from majority vote. Conclusion: The automatic detection of gender by ECG could be very important and improve the development of predictive systems for cardiovascular disease. These classifications are promising due to the use of nonstandard ECG and to the simplicity of extraction of features that potentiated the correct classification