A1041
Title: A frequency domain functional approach for time series classification with application to epileptic seizure
Authors: Kun Chen - Southwestern University of Finance and Economics (China) [presenting]
Rui Huang - Nanjing University (China)
Xingzuo He - Southwestern University of Finance and Economics (China)
Abstract: The automated diagnosis of epileptic seizures via electroencephalogram (EEG) signals poses significant challenges due to their high-dimensional nature and inherent temporal dependencies. Addressing these issues is crucial for improving diagnostic accuracy in clinical settings. A novel frequency domain functional approach is introduced, targeted at enhancing the classification of time series data for such applications. The method centers on the spectral density function, which captures the second-order dynamics of general stationary processes, utilizing log-periodogram ordinates as asymptotically unbiased estimators of the log-spectral density functions. These ordinates, approximately independent across different frequencies, are transformed into smooth curves to facilitate the application of functional principal component analysis combined with a distance-based classification rule. It is shown that the misclassification rates tend to be zero under mild conditions. Based on extensive simulations in various scenarios and a real application to EEG signals of epilepsy seizures, the efficacy of the proposed method is proven.
