“Epileptic seizure prediction” can be defined as an identification of a time or period when seizure are probably approaching without knowing when is the actual time they are going to occur (Viglione & Walsh, 1975) . It is very important to predict when seizure would happen as seizure can happen at any time in epileptic seizure patient’s daily life. A diverse methods and approaches had been done since 30 years ago to help in predicting epilepsy. Nonetheless, there are still few aspects that limit the implementation of those approaches to clinical practices. One of the problems that is still being addressed in early year of 2020 (Rasheed et al., 2021) is regarding data dimensionality that comes from multiple electrodes of EEG recordings, especially scalp EEG. This problem affects the performance of the end results. Previous works had shown the impressive results in terms of accuracy when dimensionality reduction methods are used to their models. But, how does time prediction before seizure onset changed when these methods are used? Hence, this research is proposed to explore and experiment methods to reduce the effects of curse of data dimensionality that came from scalp EEG. Using dataset obtained from CHB-MIT, ictal and pre ictal signal will be decompose using discrete wavelet transform (DWT), extract their features, reduce data dimension by comparing between Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbour Embedding (t-sne), classify Support Vector Machine (SVM) approach then compare their performance against time by using firing power concept.