The current design of neural network relies heavily on the subjective judgment and heuristic steps performed by expert architecture designers. In this research article, we suggest an automatic technique that optimize and analyze the intracranial electroencephalogram (iEEG) signal data using neural network structures. Our approach effectively decreases the human dependency and experimental mystery to advance neural models. The proposed benchmark model improved kappa score, macro F1 score, AUROC mean, and AUPRC mean corresponding to powerline, noise, pathology, and physiology event classes from 0.92 to 0.94, 0.86 to 0.91, 0.9523 to 0.9613 and 0.8246 to 0.8358 for two autonomous datasets taken from St. Anne's college hospital (Brno, Czech Republic) and Mayo clinic (Rochester, MN, United State). Our proposed method (benchmark model) achieved significantly improvement in McNemar test (a non-parametric test for paired nominal data) values when compared with other models. In benchmark model, McNemar’s test correct values increase from 45 to 48 compared to short time Fourier transform model for FNUSA and Mayo clinic dataset respectively. Whereas McNemar’s test correct values increase from 22 to 30 compared to wavelet scalogram transform model for FNUSA and Mayo clinic dataset respectively.