Скачать или смотреть Comparing SVM, DFA, and random forest for classifying semantic prediction from EEG signals - CNS2022

Comparing support vector machines, discriminant function analysis, and
random forest for classifying semantic prediction from EEG signals using
averaged- and single- trial data
Timothy Trammel1,2, Natalia Khodayari3, Matthew J. Traxler1,2, Tamara Y. Swaab1,2
1 Dept of Psychology, University of California, Davis (UCD), 2 Center for Mind and Brain, UCD, 3 Dept of Psychological and Brain Sciences, Johns Hopkins University

While conventional univariate analyses of electroencephalogram (EEG) and ERP data continue to provide valuable insights into the neural computations underlying visual word recognition, it has been shown in recent years that multivariate methods using machine-learning classification provide powerful tools to investigate the content of neural computations. EEG decoding studies commonly use support vector machines (SVM), discriminant function analysis (DFA), or random forests (RF), without justification for the classification method chosen. To our knowledge, there have been no formal performance comparisons between these three models for classifying EEG data. The present study aims to compare these models' performance while classifying EEG data from two datasets (visual prediction accuracy priming paradigm and semantic relatedness priming paradigm) to address the following questions: 1) Can SVMs, DFAs, and RFs each classify EEG data according to successful prediction or semantic relatedness? 2) Are there any significant differences between the models when classifying the EEG data? 3) If there are differences, how do the models differ in classification performance? 4) Are classifier performance differences specific to the dataset being analyzed? Permutation-based cluster analyses of the models over the time course of the data show that: 1) all three classifiers can reliably decode both prediction accuracy and semantic relatedness from the EEG signal in both datasets and 2) that the SVM classifier significantly outperforms the DFA and RF classifiers when classifying both averaged-trial data using both datasets and single-trial data using the prediction accuracy dataset. However, in the semantic relatedness priming data, the models perform similarly.