Komparasi Algoritma Machine learning untuk Klasifikasi Kondisi Mental Siswa Berdasarkan Sinyal EEG Wearable
Abstrak
The quantifying of the minds of students in objective, real-time and application of this information in education is one of the biggest issues in education in general, and in learning of numbers and mathematics in particular. In this research paper, we are going to compare the performance of five machine learning algorithms such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Decision Tree, Naive Bayes and random forest in classifying three mental states such as relaxed, neutral and concentrating using EEG signals of a low cost wearable device, the Muse headband. The data had been received in the open dataset EEG Brainwave Dataset: Mental State on Kaggle where they were recorded using four channels (TP9, AF7, AF8, TP10). The signal processing pipeline involves signal pre-processing (band-pass filter), feature extraction (mean, standard deviation, skewness, kurtosis) of both time domain and frequency domain (delta, theta, alpha and beta bandpower) and dimensionality reduction (Principal Component Analysis (PCA)). Accuracy, precision, recall and F1-score were tested on 102 test samples. They found that the most successful one was Random Forest whose accuracy was 0.84 and macro F1-score was 0.84 and SVM (0.82), KNN (0.78), Decision Tree (0.75) and Naive Bayes(0.68) followed. A constant problem in all of the models was the neutral condition because the signal characteristics of the neutral condition lie between those of the other two conditions. These results imply that a set of basic characteristics of a low-cost wearable EEG device and the random forest algorithm may already be sufficient to classify mental states, and has potential as a platform on which to build adaptive, real-time cognitive preparedness monitoring systems,
Identifier
- DOI: https://doi.org/10.51901/jiifkom.v5i02.743
- OpenAlex: https://openalex.org/W7168139014