“Preictal labeling” plays a vital role in predicting epileptic seizures in advance. Though several machine and deep learning-based seizure prediction models are developed, annotating the signals is performed based on the literature and trials which certainly degrades the performance of the learning models as the preictal state varies among the epileptic patients. Thus, this paper introduces a first-of-a-kind, exponential energy-based statistical model to identify the states of the EEG signals thereby ensuring precise preictal labeling. The proposed statistical model is evaluated by utilizing the benchmark EEG dataset obtained from CHB-MIT database in terms of false prediction rate and accuracy. The experimental results confirm that preictal labeling performed using the proposed model significantly increases the performance of the learning model.