癲癇發作會使患者喪失意識，可能導致癲癇患者受到傷害，而現有的治療方式包含藥物控制、手術以及神經調控，對於部分癲癇患者不適用，癲癇預測方法可以在發作前發出警告，使患者有時間應對來保障人身安全和生活品質。部分現有癲癇預測方法採用機率平均方法，忽略腦電圖特徵隨時間變化的性質；結合自迴歸模型特徵與傳統機器學習模型，例如最小絕對值收斂與選擇算子 (Least absolute shrinkage and selection operator, LASSO)，未特別考慮自迴歸模型係數中相鄰數值的關係，以上兩點可能使癲癇預測的準確度受限。本研究採自注意力機制 (Self-attention mechanism) 進行時序特徵融合，讓分類模型可根據不同時間窗特徵對於分類的貢獻大小，強化重要特徵並抑制較不具貢獻的特徵影響，取代機率平均方法；採用自迴歸模型結合一維卷積神經網路 (One-dimensional convolutional neural network, 1D-CNN)，捕捉自迴歸模型係數局部的衰減或振盪特徵，取代傳統機器學習模型，結合1D-CNN和自注意力機制進行癲癇預測。使用Kaggle和CHB-MIT資料集評估分類模型性能，自迴歸模型結合1D-CNN並採用機率平均方法，測試集平均AUC達0.72，優於LASSO的0.63，儘管未達顯著差異 (paired t-test, p=0.22)，仍顯示1D-CNN學習的自迴歸模型局部特徵可能有助於提升癲癇預測性能；另一方面，自注意力機制並未進一步提升癲癇預測性能，但根據注意力權重的分析發現，部分受試者的注意力權重與時間推移呈現顯著正相關，表示模型傾向對較接近發作時間的特徵給予較高權重，顯示分類模型可學習到部分腦電圖特徵的時序變化關係。本研究結果顯示，以1D-CNN學習自迴歸模型係數的局部特徵具有提升癲癇預測性能的潛力，而自注意力機制則可協助模型學習腦電圖特徵的時序變化關係。

Epileptic seizures may cause loss of consciousness and injury, while existing treatments are not effective or suitable for all patients. Seizure prediction can provide early warnings to improve patient safety and quality of life. However, existing methods may overlook the temporal variation of electroencephalographic (EEG) features when using probability averaging and may not adequately capture local relationships among adjacent autoregressive (AR) coefficients when using conventional machine learning models, such as LASSO. This study therefore combines an AR model with a one-dimensional convolutional neural network (1D-CNN) to capture local patterns in AR coefficients and employs a self-attention mechanism to adaptively integrate features across different time windows. The proposed methods were evaluated using the Kaggle and the CHB-MIT dataset, with a subject-specific model trained for each subject. The mean test AUC was 0.72 for the AR model with 1D-CNN and probability averaging, compared with 0.63 for LASSO. While the difference was not statistically significant (paired t-test, p=0.22), the results suggest that learning the local AR coefficient patterns using 1D-CNN could improve the seizure prediction performance. Self-attention did not further improve predictive performance; however, analysis of attention weights indicated that the model could learn temporal relationships in EEG features, assigning greater importance to features closer to seizure onset.

[1] R. S. Fisher et al., “Epileptic Seizures and Epilepsy: Definitions Proposed by the International League Against Epilepsy (ILAE) and the International Bureau for Epilepsy (IBE),” Epilepsia, vol. 46, no. 4, pp. 470–472, 2005.
[2] D. Bhalla, B. Godet, M. Druet-Cabanac, and P.-M. Preux, “Etiologies of epilepsy: A comprehensive review,” Expert Rev. Neurother., vol. 11, pp. 861–76, Jun. 2011.
[3] S. Ghosh et al., “Pharmacological and Therapeutic Approaches in the Treatment of Epilepsy,” Biomedicines, vol. 9, no. 5, Art. no. 5, May 2021.
[4] A. Khoo, J. de Tisi, S. Mannan, A. G. O’Keeffe, J. W. Sander, and J. S. Duncan, “Reasons for not having epilepsy surgery,” Epilepsia, vol. 62, no. 12, pp. 2909–2919, 2021.
[5] P. Ryvlin, S. Rheims, L. J. Hirsch, A. Sokolov, and L. Jehi, “Neuromodulation in epilepsy: state-of-the-art approved therapies,” Lancet Neurol., vol. 20, no. 12, pp. 1038–1047, Dec. 2021.
[6] B. H. Brinkmann et al., “Crowdsourcing reproducible seizure forecasting in human and canine epilepsy,” Brain, vol. 139, no. 6, pp. 1713–1722, Jun. 2016.
[7] A. H. Shoeb, “Application of machine learning to epileptic seizure onset detection and treatment,” Thesis, Massachusetts Institute of Technology, 2009. Accessed: Nov. 18, 2025.
[8] L. Chisci et al., “Real-Time Epileptic Seizure Prediction Using AR Models and Support Vector Machines,” IEEE Trans. Biomed. Eng., vol. 57, no. 5, pp. 1124–1132, May 2010.
[9] J. Hu, L. Shen, S. Albanie, G. Sun, and E. Wu, “Squeeze-and-Excitation Networks,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 7132–7141.
[10] L. Marsh and V. Rao, “Psychiatric complications in patients with epilepsy: a review,” Epilepsy Res., vol. 49, no. 1, pp. 11–33, Mar. 2002.
[11] O. Devinsky, T. Spruill, D. Thurman, and D. Friedman, “Recognizing and preventing epilepsy-related mortality,” Neurology, vol. 86, no. 8, pp. 779–786, Feb. 2016.
[12] L. Dalic and M. J. Cook, “Managing drug-resistant epilepsy: challenges and solutions,” Neuropsychiatr. Dis. Treat., vol. 12, pp. 2605–2616, Oct. 2016.
[13] D. Schmidt, “Drug treatment of epilepsy: Options and limitations,” Epilepsy Behav., vol. 15, no. 1, pp. 56–65, May 2009.
[14] P. J. Karoly et al., “The circadian profile of epilepsy improves seizure forecasting,” Brain, vol. 140, no. 8, pp. 2169–2182, Aug. 2017.
[15] P. Rajna et al., “Hungarian multicentre epidemiologic study of the warning and initial symptoms (prodrome, aura) of epileptic seizures,” Seizure, vol. 6, no. 5, pp. 361–368, Oct. 1997.
[16] A. Schulze-Bonhage, C. Kurth, A. Carius, B. J. Steinhoff, and T. Mayer, “Seizure anticipation by patients with focal and generalized epilepsy: A multicentre assessment of premonitory symptoms,” Epilepsy Res., vol. 70, no. 1, pp. 83–88, Jul. 2006.
[17] H. H. Lange, J. P. Lieb, J. Engel, and P. H. Crandall, “Temporo-spatial patterns of pre-ictal spike activity in human temporal lobe epilepsy,” Electroencephalogr. Clin. Neurophysiol., vol. 56, no. 6, pp. 543–555, Dec. 1983.
[18] V. Navarro, J. Martinerie, M. Le Van Quyen, M. Baulac, F. Dubeau, and J. Gotman, “Seizure Anticipation: Do Mathematical Measures Correlate with Video-EEG Evaluation?,” Epilepsia, vol. 46, no. 3, pp. 385–396, 2005.
[19] L. D. Iasemidis and J. C. Sackellares, “■ REVIEW : Chaos Theory and Epilepsy,” The Neuroscientist, vol. 2, no. 2, pp. 118–126, Mar. 1996.
[20] L. Kuhlmann, K. Lehnertz, M. P. Richardson, B. Schelter, and H. P. Zaveri, “Seizure prediction — ready for a new era,” Nat. Rev. Neurol., vol. 14, no. 10, pp. 618–630, Oct. 2018.
[21] S. Selim, E. Elhinamy, H. Othman, W. Abouelsaadat, and M. A.-M. Salem, “A Review of Machine Learning Approaches for Epileptic Seizure Prediction,” in 2019 14th International Conference on Computer Engineering and Systems (ICCES), Feb. 2019, pp. 239–244.
[22] Department of Electrical and Computer Engineering, International Islamic University of Malaysia, Kuala Lumpur, Malaysia, A. R. Najeeb, M. J. E. Salami, T. Gunawan, and A. M. Aibinu, “Review of Parameter Estimation Techniques for Time-Varying Autoregressive Models of Biomedical Signals,” Int. J. Signal Process. Syst., vol. 4, no. 3, pp. 220–225, Jun. 2016.
[23] R. A. Stefanescu, R. G. Shivakeshavan, and S. S. Talathi, “Computational models of epilepsy,” Seizure, vol. 21, no. 10, pp. 748–759, Dec. 2012.
[24] P.-N. Yu, C. Y. Liu, C. N. Heck, T. W. Berger, and D. Song, “A sparse multiscale nonlinear autoregressive model for seizure prediction,” J. Neural Eng., vol. 18, no. 2, p. 026012, Feb. 2021.
[25] S. Kiranyaz, O. Avci, O. Abdeljaber, T. Ince, M. Gabbouj, and D. J. Inman, “1D convolutional neural networks and applications: A survey,” Mech. Syst. Signal Process., vol. 151, p. 107398, Apr. 2021.
[26] O. Abdeljaber, O. Avci, S. Kiranyaz, M. Gabbouj, and D. J. Inman, “Real-time vibration-based structural damage detection using one-dimensional convolutional neural networks,” J. Sound Vib., vol. 388, pp. 154–170, Feb. 2017.
[27] S. Kiranyaz, T. Ince, and M. Gabbouj, “Personalized Monitoring and Advance Warning System for Cardiac Arrhythmias,” Sci. Rep., vol. 7, no. 1, p. 9270, Aug. 2017.
[28] 蔡明倫, “定向轉移函數基於卷積神經網路與 L1 正則化線性迴歸之癲癇預測,” 碩士論文, 國立成功大學機械工程學系, 臺灣臺南市, 2023.
[29] X. Wang, C. Zhang, T. Kärkkäinen, Z. Chang, and F. Cong, “Channel Increment Strategy-Based 1D Convolutional Neural Networks for Seizure Prediction Using Intracranial EEG,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 31, pp. 316–325, 2023.
[30] A. K. Ibrahim, H. Zhuang, E. Tognoli, A. Muhamed Ali, and N. Erdol, “Epileptic seizure prediction based on multiresolution convolutional neural networks,” Front. Signal Process., vol. 3, May 2023.
[31] F. M. Shiri, T. Perumal, N. Mustapha, and R. Mohamed, “A Comprehensive Overview and Comparative Analysis on Deep Learning Models: CNN, RNN, LSTM, GRU,” J. Artif. Intell., vol. 6, no. 1, pp. 301–360, 2024.
[32] S. Yang, X. Yu, and Y. Zhou, “LSTM and GRU Neural Network Performance Comparison Study: Taking Yelp Review Dataset as an Example,” in 2020 International Workshop on Electronic Communication and Artificial Intelligence (IWECAI), Jun. 2020, pp. 98–101.
[33] Christopher M. Bishop and Hugh Bishop, Deep Learning: Foundations and Concepts. Springer Cham, 2024.
[34] A. Vaswani et al., “Attention is All you Need,” in Advances in Neural Information Processing Systems, Curran Associates, Inc., 2017.
[35] M. Eberlein et al., “Convolutional Neural Networks for Epileptic Seizure Prediction,” in 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Feb. 2018, pp. 2577–2582.
[36] M. Sun, F. Wang, T. Min, T. Zang, and Y. Wang, “Prediction for High Risk Clinical Symptoms of Epilepsy Based on Deep Learning Algorithm,” IEEE Access, vol. 6, pp. 77596–77605, 2018.
[37] A. D. Krystal, R. Prado, and M. West, “New methods of time series analysis of non-stationary EEG data: eigenstructure decompositions of time varying autoregressions,” Clin. Neurophysiol., vol. 110, no. 12, pp. 2197–2206, Dec. 1999.
[38] X. Lun, Z. Yu, T. Chen, F. Wang, and Y. Hou, “A Simplified CNN Classification Method for MI-EEG via the Electrode Pairs Signals,” Front. Hum. Neurosci., vol. 14, Sep. 2020.
[39] D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization,” Jan. 30, 2017, arXiv: arXiv:1412.6980.
[40] X. Wei, L. Zhou, Z. Zhang, Z. Chen, and Y. Zhou, “Early prediction of epileptic seizures using a long-term recurrent convolutional network,” J. Neurosci. Methods, vol. 327, p. 108395, Nov. 2019.
[41] H. He and E. A. Garcia, “Learning from Imbalanced Data,” IEEE Trans. Knowl. Data Eng., vol. 21, no. 9, pp. 1263–1284, Sep. 2009.
[42] Y. Zhang et al., “Epileptic Seizure Detection Based on Bidirectional Gated Recurrent Unit Network,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 30, pp. 135–145, 2022.
[43] L. J. P. van der Maaten and G. E. Hinton, “Visualizing High-Dimensional Data Using t-SNE,” J. Mach. Learn. Res., vol. 9, pp. 2579–2605, 2008.