Epilepsy is a prevalent neurological disorder characterized by recurrent seizures, significantly impacting patient quality of life. Accurate seizure prediction using electroencephalogram (EEG) data has the potential to revolutionize patient care by enabling timely interventions. This study reviews the latest machine learning and deep learning advances for seizure prediction, focusing on transfer learning techniques applied to EEG signals. Among the evaluated models, VGG16 demonstrated outstanding performance, achieving 93.33% accuracy with perfect sensitivity and high specificity, highlighting its effectiveness even with limited training data. Res Net architectures showed mixed results, with ResNet101 achieving high recall and specificity but lower sensitivity, while ResNet50 underperformed in overall accuracy. Other models such as DenseNet201 and X ception exhibited lower accuracy, emphasizing the need for further tuning and pre-processing. The findings underscore the advantages of transfer learning and highlight ongoing challenges including data scarcity and model generalizability. This paper discusses strategies to overcome these barriers and outlines future research directions for clinically deployable seizure prediction systems.