| The Genetic Algorithm (GA) is used for the classification problem of Electrocardiogram (ECG), by optimizing the weights in feature extraction and selection. Most of the related approaches in GA optimized feature classification and selection focus on some particular classifier settings and the results are mostly empirical. In this paper, the feature weighting of three simple classifiers: The k-Nearest Neighbor classifier, the Bayes Classifier and the Linear Regression Classifier are tested to give a more comprehensive evaluation of GA. The utility of the feature optimization on these classifiers is analyzed analytically and is tested empirically. Some traditional feature transformation techniques, such as the Principal Component Analysis (PCA), the Linear Discriminant Analysis (LDA) and their combinations are also tested with the GA optimized feature selection/extraction. All the experiments are tested with the ECG data and other benchmark pattern classification data sets. The validation results are presented with a two-tailed T test of 95% significance interval, showing that GA is efficient in finding the optimal feature weighs from the training samples, thus improving the performance of the validation patterns in most cases. In order to extract feature points from the ECG, a simple ECG preprocessing algorithm is proposed and 21 morphological features and 2 frequency features of the ECG signal have been extracted. |
