Epileptic seizures are neurological events that can be effectively monitored using electroencephalography (EEG). Early and accurate detection of seizures plays a crucial role in patient safety and diagnosis, yet manual interpretation of electroencephalography data remains time-consuming and prone to variability. In this study, we propose a convolutional neural network-based model trained on the CHB-MIT Scalp electroencephalography dataset to automatically classify electroencephalography segments as seizure or non-seizure. Given the extreme class imbalance typical in seizure datasets, we introduce class weighting during training to improve sensitivity to seizure events without severely compromising specificity. The model achieves high overall performance with a validation accuracy of 99.0%, recall of 94.7%, and a precision of 24.7% for seizure detection. We also present a comparative analysis of the model’s behavior with and without class weighting, showing a notable improvement in true positive detection at the cost of a manageable increase in false positives. These results demonstrate the effectiveness of convolutional neural network in capturing spatial-temporal patterns in electroencephalography data and highlight the importance of imbalance-aware training strategies for real-world seizure detection applications.