A Comparative Analysis of Preprocessing Methods for Single-Trial Event Related Potential Detection

摘要

The choice of a suitable preprocessing method for single-trial event-related potential (ERP) data has fundamental importance because it may improve the efficiency of a brain-computer interface (BCI) system. However, the selection of an appropriate method can be challenging and may depend on the type of data as well. In order to elaborate on this point, this manuscript investigates the impact of preprocessing on single-trial ERP detection. Method: This manuscript has investigated three scenarios for preprocessing ERP data: (1) ERP analysis without any preprocessing; (2) ERP analysis involving the amplitude-based artifact rejection method; (3) ERP analysis based on a combination of the amplitude-based artifact rejection and wavelet-enhanced independent component analysis (wICA) methods. In particular, the comparison of these preprocessing methods was based on a common machine learning (ML) framework. Therefore, the three different preprocessing methods could be compared. Also, three different classifiers (i.e., logistic regression (LR), k-nearest neighbors (kNN) and support vector machine (SVM)) were compared as well. In addition, the performance metrics utilized for this purpose were the single-trials classification recognition, precision and recall. The proposed ML framework involved the general sub-blocks of feature extraction, feature selection, classification and validation (10-fold cross-validation). The proposed ML framework was aimed to classify the single trials event-related potentials (ERPs) such as P300 (target) vs non-P300 (non-target). Results: The maximum classification accuracy was achieved for scenario 3 as described above, i.e., the combination of amplitude-based artifact rejection and wICA methods. More specifically, the classification results were as follows: recognition = 0.77, recall = 0.77, and precision = 0.99. Conclusion: The choice of an EEG preprocessing method has a significant impact on the subsequent analysis.