Optimal EEG Time Window Length for Boredom Classification Using Combined Non-Linear Features

In this paper, we investigate the effect of different time window lengths on boredom detection to find the optimal length for extracting EEG features related to boredom. Continuous EEG was recorded from 9 healthy participants (mean age = 31.8, 5 female) as they watched an education video stimulus to induce boredom. A range of non-linear features, such as fractal dimension, entropy, and chaotic features, were computed over time windows of 0.5, 1, 2, 3, 4, 5, 6, 10, 15, 20, 25, and 30 seconds. Feature combination was achieved with the help of feature selection using the Gini impurity score. Classifiers, namely, XG Boost, random forest (RF), and multilayer perceptron were employed to assess boredom classification accuracy systematically and compared the signal features performance calculated in different time windows. The proposed framework yields the highest mean accuracy of 88.61%±0.10 with 5-fold cross-validation (CV) and leave-one-subject-out (LOSO) CV accuracy of 88.92%±0.25 when using a 0.5-sec window length based on the combined feature set and RF classifier. These results provide a clear reference for selecting epoch duration when analyzing EEG signals related to boredom, an important methodological consideration for any EEG-driven emotion recognition systems.