This paper presents a frame work for the segmentation of EEG signals into three distinctive patterns; normal, pre-ictal and ictal based on sub-band decomposition. The objective of this segmentation process is to implement it on a mobile connected wirelessly to the electrode headset in order to give audio or visual alarms to epilepsy patients in case of epileptic seizure approaching. EEG signals contain five bands; Delta, Theta, Alpha, Beta, and Gamma (δ, θ α, β, and γ). The study in this paper tests each sub-band for possibility of seizure prediction. The sub-band decomposition is performed with IIR filters. The prediction method adopts a statistical approach that has training and testing phases. The training phase comprises estimation of five signals attributes; amplitude, derivative, local mean, local variance, and median. The PDF of each attribute is estimated for normal and pre-ictal states. Based on pre-set prediction probability and false alarm probability constraint a process of channel selection and bin selection from the PDFs of the selected as a tool for feature reduction and selection. The testing phase is performed with a threshold strategy on the selected bins. A majority voting strategy with a moving average smoothing filters is used for decision making. Simulation results proved the feasibility of the gamma band for seizure prediction.