Studi Komparasi Algoritma Klasifikasi Mental Workload Berdasarkan Sinyal EEG

Dessy Kusumaningrum; Imah, Elly Matul · 2020 · Crossref

DOI: 10.37396/jsc.v3i2.69

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of classifying mental workload levels using Electroencephalogram (EEG) signals, motivated by the need to monitor cognitive fatigue that impacts productivity and safety. Excessive mental workload leads to decreased performance, making real-time detection crucial. The research aims to identify the most effective classification algorithm by comparing Random Forest, K-Nearest Neighbor (KNN), Multi-Layer Perceptron (MLP), and Support Vector Machine (SVM) based on accuracy, model building time, and memory usage. The experimental design utilized the STEW (Simultaneous Task EEG Workload) dataset, comprising data from 45 male subjects performing multitasking tasks. The study focused on the AF3 EEG channel, selected for its high accuracy in cognitive workload classification. Data preprocessing involved segmenting raw signals into 64 chunks per subject, resulting in a matrix of 5760 instances with 300 attributes. Feature extraction was performed using Principal Component Analysis (PCA), reducing the dimensionality to 23 principal components. Classification was conducted using 10-fold cross-validation (noted as 100-fold in one section, but 10-fold in the conclusion) via the Weka software, while preprocessing and feature extraction were handled in MATLAB. The mental workload was categorized into three classes: low, moderate, and high. The results indicate that KNN achieved the highest classification accuracy at 70.161%, followed by Random Forest at 63.640%. MLP and various SVM configurations (Linear, Polykernel, RBF) performed poorly, with accuracies ranging from 42.532% to 51.002%. In terms of computational efficiency, KNN required zero time for model building as it is a lazy learner, but it demanded significantly higher memory usage to store the entire training dataset. Conversely, Random Forest required minimal time to build the model and used less memory than KNN. SVM algorithms, particularly the Polykernel variant, had the longest model building times and high memory consumption due to kernel matrix storage. The study concludes that while KNN offers the best accuracy, Random Forest is the recommended algorithm for real-time applications due to its balance of reasonable accuracy, fast model construction, and lower memory requirements. The findings suggest that tree-based and instance-based learners outperform neural networks and support vector machines for this specific EEG-based mental workload classification task when using PCA-extracted features.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms