An Efficient Optimized Feature Selection with Machine Learning Approach for ECG Biometric Recognition

In machine learning, an efficient classifier model design is mostly based on effective feature extraction and appropriate feature selection. This work mainly focused on different optimized feature selection algorithms for automatic biometric recognition system with the use of electrocardiogram (ECG) signals. Initially, the features are extracted from P-QRS-T segments of the ECG signal with position normalization. The extracted features are processed through different optimization algorithms for quality assessment prior to the classification stage. In this work, two different methods are proposed based on wrapper feature selection and embedded feature selection to get optimized feature datasets for the perfect identification of human biometrics. In the first proposed method, wrapper-oriented feature selection is implemented with genetic algorithm (GA) and particle swarm optimization. In the second proposed method, the embedded method is included as the least absolute shrinkage selection operator and elastic net (EN). The processed optimized feature vectors from the optimization phase are fed to popular machine learning techniques, such as support vector machine and random forest (RF) classifiers, for automatic biometric human recognition. Classifier performances are investigated using two publicly available open-source databases, Electrocardiogram identification and large Physikalisch-Technische Bundesanstalt diagnostic database with a dynamic number of subjects. The experimental results indicate that the proposed combination of feature selection with machine learning algorithms has enhanced the classification accuracy. Finally, GA and EN with RF classifier provide improved recognition rates of 95.30% and 94.90%, respectively.