This paper introduces KAN-EEG, a novel architecture for seizure detection using EEG signals, aiming to replace traditional MLP-based models. The KAN-EEG model is tested on three diverse datasets from the USA, Europe, and Oceania, demonstrating strong generalization across different geographical regions. The model uses learnable activation functions and efficient data representation, enabling it to handle complex EEG data effectively. Compared to traditional MLPs, KANs show better performance with smaller network sizes, making them more computationally efficient and suitable for real-time applications. The study highlights the KAN's adaptability, robustness, and efficiency in seizure detection, with results showing comparable or superior performance to existing models. The KAN-EEG model was trained on 400 hours of EEG data, achieving an AUROC of 0.89, and demonstrated robustness in both in-sample and out-of-sample testing. The model's ability to generalize across different datasets and its efficiency in processing EEG signals make it a promising alternative to traditional MLPs for seizure detection. The study also emphasizes the potential of KANs in medical diagnostics, offering a more accurate, efficient, and interpretable tool for seizure detection. The results indicate that KANs can replace MLP-based models for seizure detection, providing a more effective solution for clinical applications. The study concludes that KANs have significant potential in improving clinical outcomes for patients with epilepsy by offering a more accurate and efficient tool for seizure detection.This paper introduces KAN-EEG, a novel architecture for seizure detection using EEG signals, aiming to replace traditional MLP-based models. The KAN-EEG model is tested on three diverse datasets from the USA, Europe, and Oceania, demonstrating strong generalization across different geographical regions. The model uses learnable activation functions and efficient data representation, enabling it to handle complex EEG data effectively. Compared to traditional MLPs, KANs show better performance with smaller network sizes, making them more computationally efficient and suitable for real-time applications. The study highlights the KAN's adaptability, robustness, and efficiency in seizure detection, with results showing comparable or superior performance to existing models. The KAN-EEG model was trained on 400 hours of EEG data, achieving an AUROC of 0.89, and demonstrated robustness in both in-sample and out-of-sample testing. The model's ability to generalize across different datasets and its efficiency in processing EEG signals make it a promising alternative to traditional MLPs for seizure detection. The study also emphasizes the potential of KANs in medical diagnostics, offering a more accurate, efficient, and interpretable tool for seizure detection. The results indicate that KANs can replace MLP-based models for seizure detection, providing a more effective solution for clinical applications. The study concludes that KANs have significant potential in improving clinical outcomes for patients with epilepsy by offering a more accurate and efficient tool for seizure detection.