According to WHO, 6 million people are affected by an epileptic seizure every year as per a survey carried out in 2019. At the moment, doctors use direct observation of the electroencephalogram (EEG) signal to determine the presence of an epileptic seizure. However, epileptic detection in most of the previous research works suffers from low accuracy and is unsuitable for processing large datasets. In this work, the seizure EEG signal is effectively detected and enhanced using Chebyshev normalization. Additionally, the signals are decomposed by applying fast empirical mode decomposition (EMD). Then, entropy features are extracted and effective selection is obtained by using the improved artificial bee colony (ABC) optimization algorithm. Finally, a stacked autoencoder (SAE) is used for better EEG classification. The existing researches such as MCAFF, CNN-RNN, ESSA, SVM and 1D-CNN are used for comparing the IABC-SAE method. The proposed IABC-SAE method gained better performance in seizure EEG signal identification and achieves higher classification accuracy (CA) of 99.98% in TUH-EEG database compared to the existing ESSA.