PSO–GA Hybrid Feature Optimization for Cardiac Disease Detection Using ECG Signals and Machine Learning Classifiers

Among the most life-threatening diseases in the world is the cardiovascular disease and detection of such disease at an appropriate time and with the right diagnosis is crucial in reducing the risks of such disease in the clinical setting and increase the chance for positive results in the patients. In cardiac diagnosis, information regarding electrical activity of the heart can be collected at a standard way, which is the Electrocardiogram (ECG) signals and collected from the surface of the body without invasiveness. But, the detection of this using the ECG is prone to noise in ECG, irrelevant features, class imbalance and poor selection of model parameters. The problems can affect the correctness of automation in the diagnosis. This study is to design hybrid Particle Swarm Optimization (PSO) Approach with Genetic Algorithm (GA) for detection of cardiac disease using ECG. For this purpose, PSO–GA algorithm is introduced for feature selection and to optimize the performance of the classifiers. Particle Swarm Optimization and Genetic Algorithm can be used to rapidly traverse the entire world and diversify the solutions through selection, crossover and mutation. Three machine learning classifiers Support Vector Machine, Random Forest and AdaBoost are used for the classification of the optimized subset of features. The benchmark datasets: MIT-BIH Arrhythmia and PTB-XL are just used for the purpose of testing the model in different cases of ECG classification. A wide range of experimental results distinctly shows that it is better to use the PSO–GA based optimized classifiers as compared to all the non-optimized classifiers. Among the three classifiers, Random Forest with PSO – GA could achieve the highest accuracy score and F1 score with both the data sets.