We investigate the detection of epileptic seizure onset based on electroencephalography (EEG) signal in short-time sessions (less than one second) from various samples from epilepsy and healthy people. Wavelet transform methods are applied to extract the features embedded in the high-dimensional epileptic EEG signal. It is found that results of wavelet transform play significant roles in dimensional reduction process. Then, machine learning pipeline is built based on support vector machine (SVM) algorithm. It is found that epileptic seizure state in the test data set could be predicted with high precision (98.1%) based only on mini-segments of EEG signal (0.6 second). Predictions based on 0.1 second mini-segments of EEG signal are also investigated. This research may be significant to the clinical treatment of epileptic seizure, because efficient methods could be applied to interrupt the process of epileptic seizure very fast (in less than one second).