Math game performance prediction from EEG signals using machine learning algorithms

The study aimed to predict game performance by using EEG (Electroencephalography) data obtained while playing a digital game designed for the mathematics curriculum. For this purpose, an educational digital mathematics game developed within the scope of the study was played to the participants and their brain signals were recorded simultaneously. The participants are male volunteers between 18 and 25, selected by convenient sampling method from computer engineering students. The data obtained from the participants were combined, game scores were calculated and labeled, and the data pre-processing process began. During the data pre-processing process, missing data were removed and noise was reduced by applying normalization and outlier detection processes. In the feature selection process, the best discriminant features were determined by using the X² (chi-square) function. The features that best differentiate game performance are POW.FC5.Gamma, POW.AF3.Gamma, EEG.O1, POW.P8.Gamma and POW.AF3.BetaH, respectively. These attributes reflect cognitive processes such as attention, visual processing and spatial awareness and are directly related to game performance. In the study, various machine learning algorithms such as k-Nearest Neighbors (kNN), Support Vector Machines (SVM), Random Forest (RF), AdaBoost, Gradient Boosting, Neural Networks (NN) and Stochastic Gradient Descent (SGD) were used to predict game performance. The performance of the algorithms was measured by the AUC (Area Under the Curve) value and the Gradient Boosting algorithm reached the highest accuracy with 87.7%. As a result of these findings, the best distinguishing features of the EEG data were identified and it was observed that these features directly reflected the cognitive processes, attention levels and visual-motor skills of the players. Thus, an innovative method has been presented to evaluate students' performance in educational digital games more accurately and reliably, and significant contributions have been made to the use of neuroscientific data in education. Keywords: Educational digital game, mathematics education, machine learning, EEG (electroencephalography), BCI (brain computer interface), artificial intelligence