MSCK-Net: A Hybrid Multi-scale CNN–KAN–CatBoost Framework for EEG-based Autism Spectrum Disorder Classification

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that affects communication, social interaction, and behavioral patterns.Electroencephalography (EEG) signals have been widely employed for ASD diagnosis because they provide valuable information regarding brain activity.However, the complex, nonlinear, highdimensional, and noisy nature of EEG signals makes accurate ASD classification a challenging task.To address these challenges, this study proposes MSCK-Net, a hybrid deep learning (DL) framework that integrates Multi-Scale Convolutional Neural Networks (Multi-Scale CNNs), Kolmogorov-Arnold Networks (KANs), and CatBoost classification for EEG-based ASD classification.Initially, EEG features are normalized using Min-Max normalization to ensure consistent feature scaling.Subsequently, Multi-Scale CNN modules extract discriminative EEG representations at different temporal scales, while the KAN module learns nonlinear relationships among the extracted features.Finally, CatBoost performs ASD classification using the learned feature representations.The proposed framework was evaluated using the University of Sheffield EEG Dataset under a subject-independent evaluation protocol.Experimental results demonstrated that the proposed MSCK-Net framework achieved segment-level 99.45% precision, 99.50% F1-score, 99.46% accuracy, and 99.56% recall, while subject-level majority voting correctly classified all 17 held-out subjects (100% accuracy), achieving competitive performance compared with recent EEGbased ASD classification methods while outperforming the evaluated CNN, CNN-LSTM, CNN-GRU, and EEGNet baselines under the adopted experimental protocol.Furthermore, MSCK-Net required 31.2 ms execution time, 5.4 MB model size, and 52.8 MB memory usage, demonstrating a moderate performance-complexity trade-off.Noise analysis under different SNR conditions further assessed the framework's stability.The results demonstrate accurate and robust EEG-based ASD classification under the adopted experimental conditions.

Authors

Institutions

Publication Details

Journal
International journal of intelligent engineering and systems
Published
2026-11-17
DOI
https://doi.org/10.22266/ijies2026.1231.29
Primary Topic
EEG and Brain-Computer Interfaces
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

MSCK-Net: A Hybrid Multi-scale CNN–KAN–CatBoost Framework for EEG-based Autism Spectrum Disorder Classification

Sundus D. Hasan, Aqeel H. Al-Fatlawi, Suad Khairi Mohammed, Hanan A. R. Akkar
2 links
International journal of intelligent engineering and systems
EEG and Brain-Computer Interfaces
article

MSCK-Net: A Hybrid Multi-scale CNN–KAN–CatBoost Framework for EEG-based Autism Spectrum Disorder Classification

Sundus D. Hasan, Aqeel H. Al-Fatlawi, Suad Khairi Mohammed, Hanan A. R. Akkar
article en

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that affects communication, social interaction, and behavioral patterns.Electroencephalography (EEG) signals have been widely employed for ASD diagnosis because they provide valuable information regarding brain activity.However, the complex, nonlinear, highdimensional, and noisy nature of EEG signals makes accurate ASD classification a challenging task.To address these challenges, this study proposes MSCK-Net, a hybrid deep learning (DL) framework that integrates Multi-Scale Convolutional Neural Networks (Multi-Scale CNNs), Kolmogorov-Arnold Networks (KANs), and CatBoost classification for EEG-based ASD classification.Initially, EEG features are normalized using Min-Max normalization to ensure consistent feature scaling.Subsequently, Multi-Scale CNN modules extract discriminative EEG representations at different temporal scales, while the KAN module learns nonlinear relationships among the extracted features.Finally, CatBoost performs ASD classification using the learned feature representations.The proposed framework was evaluated using the University of Sheffield EEG Dataset under a subject-independent evaluation protocol.Experimental results demonstrated that the proposed MSCK-Net framework achieved segment-level 99.45% precision, 99.50% F1-score, 99.46% accuracy, and 99.56% recall, while subject-level majority voting correctly classified all 17 held-out subjects (100% accuracy), achieving competitive performance compared with recent EEGbased ASD classification methods while outperforming the evaluated CNN, CNN-LSTM, CNN-GRU, and EEGNet baselines under the adopted experimental protocol.Furthermore, MSCK-Net required 31.2 ms execution time, 5.4 MB model size, and 52.8 MB memory usage, demonstrating a moderate performance-complexity trade-off.Noise analysis under different SNR conditions further assessed the framework's stability.The results demonstrate accurate and robust EEG-based ASD classification under the adopted experimental conditions.

International journal of intelligent engineering and systemsVol. 19(12)
University of Technology - Iraq (IQ)
EEG and Brain-Computer Interfaces
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon