Letter Identification in EEG Signals Using Scalograms

Considering the improvements in Brain-Computer interface (BCI) systems and their cost reduction, a new chance for a normal life appeared for people with a wide range of disabilities. With the rise of artificial neural networks (ANN), traditional machine learning (ML) techniques heavily decreased in popularity. These methods are usually a good alternative and, also, an easier-to-understand way to get into data science in plenty of domains. Considering the use of scalograms as training material, the current paper proposes the utilization of classical ML algorithms, such as Random Forest (RF), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA) and k-Nearest Neighbors (KNN). The images are obtained by transforming the electroencephalogram (EEG) signals using continuous wavelet transform (CWT) in order to show their strengths in terms of learning. To illustrate the advantages of the proposed approach, a comparison of the obtained results with those obtained from training on EEG signals directly has been performed. Moreover, another comparison has been made between the results obtained with a neural network-based solution applied to both representations of the training data.