Dynamic reconstruction of electroencephalogram data using RBF neural networks

Frontiers in Neuroscience, Journal Year: 2025, Volume and Issue: 19

Published: March 28, 2025

Electroencephalography (EEG) is widely used for analyzing brain activity; however, the nonlinear and nature of EEG signals presents significant challenges traditional analysis methods. Machine has shown great promise in addressing these limitations. This study proposes a novel approach using Radial Function (RBF) neural networks optimized by Particle Swarm Optimization (PSO) to reconstruct dynamics extract age-related characteristics. recordings were collected from 142 participants spanning multiple age groups. Signals preprocessed through bandpass filtering (1-35 Hz) Independent Component Analysis (ICA) artifact removal. network was trained on time-series data with PSO employed optimize model parameters identify fixed points reconstructed system. Statistical analyses including ANOVA Kruskal-Wallis tests performed assess differences fixed-point coordinates. The RBF demonstrated high accuracy signal reconstruction across different frequency normalized root mean square error (NRMSE) 0.0671 ± 0.0074 Pearson correlation coefficient 0.0678. Spectral time-frequency confirmed s capability accurately capture oscillations. Importantly coordinates revealed distinct age-related. These findings suggest that can serve as quantitative markers aging providing new insights into age-dependent changes dynamics. proposed method offers computationally efficient interpretable potential applications neurological diagnosis cognitive research.

Language: Английский

Advances to IoT security using a GRU-CNN deep learning model trained on SUCMO algorithm

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 12, 2025

The rapid expansion of the Internet Things (IoT) has significantly improved various aspects our daily life. However, along with its benefits, new security threats such as Denial Service (DoS) attacks and Botnets have emerged. To adopt this technology integrity IoT environment, detection become crucial. This paper proposes a hybrid deep learning model that combines Convolutional Neural Network (CNN) Gated Recurrent Units (GRUs) to classify threats. CNN is used extract spatial features from network data, where on other hand GRUs for capturing temporal dependencies. combination makes effective at analysing both static dynamic data. Further, optimize performance proposed model, self-upgraded Cat Mouse Optimization (SUCMO) algorithm employed, state art optimization technique. SUCMO fine-tunes model's hyperparameters improve classification accuracy. evaluated through experiments two different datasets i.e., UNSW-NB15 BoT-IoT, results demonstrates work outperforms traditional well works.

Language: Английский