Nowadays, electrocardiogram (ECG) stands as a versatile tool, frequently used for the prediction of epileptic seizures. In this respect, various approaches have been advanced to deal with drug-resistant epilepsy. In regard to the present study, a new automatic seizure prediction method is proposed. Accordingly, the features extraction step is performed directly from the ECG signals. Subsequently, a stacked sparse auto-encoder, along with SoftMax classifier are applied for seizures prediction by classifying the background activity and the pre-ictal phase of seizures. Our proposal is evaluated through a publicly available database, involving seven patients with ten epileptic seizures. The reached results turn out to record high performance in terms of accuracy, sensitivity and specificity, which distinguishes our ECG-signal based seizure prediction approach as a promising valuable design liable to facilitate the patient’s life significantly.