Introduction: Performance estimates in automated EEG seizure classification depend on label fidelity, temporal partitioning, preprocessing placement, and correct model implementation. This controlled single-record methodological study quantified the influence of these factors and was not designed to estimate patient-independent clinical performance.

Materials and Methods: A de-identified single-channel CHB-MIT record (chb01_26; 256 Hz) was analyzed. A scripted seizure crop was compared with the official 1862-1963 s annotation. Three 100-s interictal intervals and the official ictal interval were divided into 10-s temporal blocks. Level-4 wavelet statistics were classified using Gaussian naive Bayes, radial-basis-function support vector machine, and a first-order Takagi-Sugeno-Kang system. Random-window, block-preserving, and chronological evaluations used fold-wise preprocessing. An additional contiguous-block evaluation introduced a 30-s temporal exclusion buffer, and TSK rule-count sensitivity was evaluated for 3, 5, 7, and 9 rules. A factorial analysis compared db4 and sym4 wavelets, 1-, 2-, and 4-s windows, and 0% or 50% overlap.

Results: The scripted seizure segment began at 1818.4 s, and 44.3% of its samples preceded the official onset. The reconstructed dataset contained 360 windows in 40 temporal groups. Under blocked validation, Gaussian naive Bayes, support vector machine, and Takagi-Sugeno-Kang models achieved Matthews correlation coefficients (MCCs) of 0.793, 0.784, and 0.827, respectively. A 30-s temporal exclusion buffer caused marked MCC reductions, while five TSK rules had the highest mean blocked MCC among 3, 5, 7, and 9 rules tested. The highest descriptive preprocessing MCC was 0.882 (sym4, 4-s windows, 50% overlap).

Conclusion: Annotation correction, temporal-group preservation, and explicit fuzzy-model specification materially changed within-record performance estimates. The findings constitute a single-record methodological case study; multi-record patient-independent validation remains necessary before clinical generalization.