Epilepsy affects over 65 million people worldwide, making it one of the most prevalent chronic neurological conditions, ranking fourth behind migraine, Alzheimer’s disease, and stroke. Despite advancements in anti-epileptic medications, approximately one-third of patients experience uncontrolled seizures, contributing to a 40% mortality rate in high-risk cases due to sudden unexpected death in epilepsy, accidents, or neurological complications. Current wearable technologies lack reliable real-time seizure detection, creating a critical need for systems that can predict seizures with low latency and seamless integration across devices. This paper introduces Epipredict, a wearable system designed to predict the onset of epileptic seizures using scalp electroencephalogram (EEG) signals. Epipredict integrates a smartwatch to process EarEEG signals, perform low-latency analysis, and issue vibration alerts to users while notifying caregivers via a mobile app. The system employs cloud-based computational offloading to ensure efficient processing across diverse devices, enabling neurologists to access detailed reports for improved monitoring and management, thus reducing epilepsy-related fatalities. Performance evaluation using cross-validation on the Physionet.org SIENA Scalp EEG database demonstrates that Epipredict’s performance is improved by synchronization measures and Machine Learning threshold (MLTH) feature selection, combined with machine learning classifiers. The system mitigates class imbalance in seizure data using techniques like synthetic minority oversampling (SMOTE), achieving a precision of 98%, an F-score of 99%, and high specificity, with a fast predictive time of 0.3945 s. By addressing challenges such as latency, computational efficiency, and device interoperability, Epipredict offers a robust solution for real-time seizure prediction with enhanced wearable usability and extended prediction horizons. Future work will focus on larger datasets, extended monitoring, and additional sensors to address limitations in real-world trigger variability, sensor noise, and deployment challenges.