Machine Learning Based ERP P300 Classification in EEG Signals from Online Game Player Performing GNAT Test

Online Game Disorder (OGD) is a behavioral disorder characterized by excessive online gaming, which disrupts an individual's daily life. Adolescents demonstrate a correlation between OGD and depression, as frequent gaming coincides with increased depressive symptoms. This research utilizes the Electroencephalography (EEG) method, which records electrical activity in the brain. Adolescents diagnosed with OGD will have their brain signals recorded, and the resulting EEG signals will undergo analysis. The analysis employs the Event-Related Potential (ERP) technique, explicitly targeting the P300 component, which occurs within a 200-500 ms window following the presentation of a stimulus from Go/No Go Association Task (GNAT). Classification is conducted using machine learning by comparing two machine learning models and three feature orders. The classification methods used are the random forest and decision tree models. The classification process involves two labels: the presence of the P300 ERP and the absence of the P300 ERP. These labels are inputted into the machine learning models for classification, with the resulting output used to compare the performance of the two models across different feature orders. The highest accuracy was achieved using the decision tree model with a combined feature set of first-order and second-order statistics, resulting in an accuracy of 92.42%. Performance parameters support this result, specifically with a precision of 78%, recall of 93%, and an F1-score of 85%. Although the decision tree model achieved the highest accuracy, the random forest model demonstrated more stable results.