Epilepsy, one of the most prevalent neurological disorders, requires innovative methods for seizure anticipation. This study characterizes preictal cardiac dynamics through analysis of 3,000 RR intervals ( 45 minutes) preceding seizures in 47 patients, employing phase space reconstruction. We segmented 647 time series into three segments (1,000 intervals each) and applied nonlinear techniques to identify evolving patterns. Using Takens' theorem, we determined optimal parameters: time delay (τ = 50, 90, 150) through autocorrelation analysis and embedding dimension (m = 6) via the false nearest neighbors method. Dimensionality reduction (PCA and LDA) revealed progressive separability between segments, quantified using the Fisher Discriminant Ratio (FDR). The analysis demonstrated significant differences in dynamic structure: FDR values increased from 0.329 (Segments 1-2) to 101.566 (Segments 1-3) in the reconstructed space, highlighting preictal autonomic transition. The study confirmed that longer time delays (τ = 150) maximize discrimination, revealing progressive alterations in heart rate variability. These findings support the potential of nonlinear heart rate dynamics as a preictal biomarker, while proposing a methodological framework for future seizure prediction systems based on accessible, low-cost physiological signals.