Error Detection in P300 Speller Device Applying Differential Entropy Features and Machine Learning Approaches

Spelling error is a habitual issue in the P300-speller device. The main concern is to detect the error in spelling task. During the spelling task Event Related Potential (ERP) based P300 signal is elicited. In this study a well-known approach is introduced to analysis the P300-ERP signal to detect the spelling error. A publicly available BCI NER 2015 Kaggle Competition dataset was used to analyze the signal. P300 ERP responses were recorded from 10 healthy subjects during the spelling task. After preprocessing the raw data's differential entropy, a statistical model was introduced to extract features. K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Logistic Regression (LR), and Random Forest (RF) algorithms were applied for the positive and negative feedback classification. The present research indicated that SVM achieved highest accuracy among all mentioned algorithms. SVM showed 80.48% accuracy with the differential entropy feature. Confusion matrices evaluated the performances of the classifiers.