CLASSIFYING ATTENTION STATES FROM BRAIN SIGNALS USING CONVOLUTIONAL NEURAL NETWORK (CNN)
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Malaysia Sarawak
Abstract
Description
The ability to monitor and classify attention states has profound implications in areas like education, healthcare, workplace productivity, and transportation safety. This study explores the classification of human attention states from brain signals using convolutional neural networks (CNNs) with a focus on distinguishing between focused and distracted states under two environmental conditions. EEG (Electroencephalography) signals were employed as the primary data source as it offers insights into neural patterns associated with cognitive states. EEG signals in the alpha band (8–13 Hz) were recorded from 45 healthy adults under both quiet and simulated noisy (70 dB) conditions. After preprocessing including band-pass filtering, artifact correction, and segmenting signals into 21-channel, one-second epochs, the data were fed into a three-layer convolutional neural network (CNN) featuring successive convolution, max-pooling, batch normalization, and dropout operations. The network converged into 37 epochs and achieved 91% accuracy, 93% precision, 91% recall, and a 0.92 F1‑score. Comparative benchmarks against k-Nearest Neighbors (76% accuracy) and Support Vector Machine with PCA (87% accuracy) demonstrated CNN’s superior consistency, reduced error rates, and robust feature learning. Detailed evaluation via confusion matrices and learning curves provided insights into class-specific performance and model convergence behavior. These results confirm the feasibility of real-time EEG-based attention monitoring and establish a reproducible implementation and evaluation pipeline for future cognitive state classification research.