EEG-based machine learning model for personalized power nap identification in Brain-Computer Interface

Handayani, Dini Oktarina Dwi and Yaacob, Hamwira Sakti and Agastya, I Made Artha and Suryady, Zeldi and Hanizam, Amirul Hilmi and Azmin, Aliff Azmeer Hakim (2026) EEG-based machine learning model for personalized power nap identification in Brain-Computer Interface. In: 10th International Conference on Information and Communication Technology for the Muslim World, ICT4M 2025, 26-27 November 2025, Kuala Lumpur.

	PDF - Published Version Restricted to Registered users only Download (973kB) | Request a copy
Preview	PDF - Supplemental Material Download (199kB) | Preview

Abstract

Electroencephalography (EEG) based Brain Computer Interface (BCI) systems provide a non-invasive means for monitoring sleep and cognitive states. Power naps, typically lasting between 10 and 30 minutes, are known to enhance alertness, memory, and cognitive recovery; however, their effectiveness varies widely across individuals due to differences in neural activity patterns. Current nap detection approaches often rely on generalized sleep staging or subjective reports, limiting their reliability and personalization. This study proposes a machine learning framework for personalized power nap identification using EEG signals from the Sleep-EDF Expanded Database. Two bipolar EEG channels (Fpz–Cz and Pz–Oz) sampled at 100 Hz were preprocessed with bandpass filtering (0.5– 49 Hz), segmented into 30-second epochs, and transformed into feature vectors combining statistical descriptors (mean, standard deviation, skewness, kurtosis, RMS) and spectral powers (total PSD and band-specific δ, θ, α, β, γ). The features were standardized and used to train a Multi-Layer Perceptron (MLP) model with class-weighted loss to address label imbalance. Experimental results demonstrated an overall accuracy of 94% on test data, with strong precision, recall, and F1-scores across Wakefulness, Light Sleep (N1+N2), Deep Sleep (N3+N4), and Rapid Eye Movement (REM). The confusion matrix revealed occasional misclassifications between light sleep and wakefulness, reflecting their overlapping EEG patterns. Training and validation curves confirmed stable convergence without overfitting. The model required <15 ms per segment for inference, indicating feasibility for real-time deployment on modest hardware. In summary, the proposed MLP framework demonstrates high accuracy, strong generalization, and low computational cost, making it a promising foundation for adaptive, real-time power nap detection. Future work will focus on integrating this system into a closed-loop BCI capable of detecting light sleep onset and optimizing nap timing to maximize recovery while avoiding sleep inertia.

Item Type:	Proceeding Paper (Other)
Uncontrolled Keywords:	EEG, BCI, MLP, Power Nap Detection, Sleep Stage Classification, Machine Learning, Personalized Rest, Cognitive Recovery
Subjects:	Q Science > QA Mathematics > QA76 Computer software
Kulliyyahs/Centres/Divisions/Institutes (Can select more than one option. Press CONTROL button):	Kulliyyah of Information and Communication Technology Kulliyyah of Information and Communication Technology Kulliyyah of Information and Communication Technology > Department of Computer Science Kulliyyah of Information and Communication Technology > Department of Computer Science
Depositing User:	Dr Dini Handayani
Date Deposited:	05 May 2026 16:03
Last Modified:	05 May 2026 16:03
Queue Number:	2026-04-Q3050
URI:	http://irep.iium.edu.my/id/eprint/128626

Actions (login required)

View Item