Background: Speeding violations are a primary cause of accidents on highways, leading to significant human, economic, and social consequences. While numerous studies have explored the prediction of traffic violations and accidents, few have focused on hourly violation counts using hybrid deep learning architectures. Methods: This study introduces a novel hybrid deep learning model combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks to predict the hourly number of speeding violations. The model was developed using hourly traffic data, including variables such as vehicle count, average speed, distance violations, and violation occurrence times. Following preprocessing, the dataset was divided into 70% training, 15% validation, and 15% testing sets. A 5-fold cross-validation approach was employed to assess model performance. Results: The cross-validation results demonstrated satisfactory performance in predicting speeding violations. During training, RMSE values ranged from 69.05 to 201.5, with the first fold showing anomalous performance; the remaining folds exhibited better results, with R² values between 0.68 and 0.85 and KGE values between 0.67 and 0.83. In the testing phase, average RMSE, R², and KGE were 89.4, 0.77, and 0.78, respectively, closely aligning with training phase averages, indicating robust model performance. In the validation phase, the model achieved an RMSE of 88.4, R² of 0.76, and KGE of 0.79. Conclusion: The proposed hybrid deep learning model effectively identifies high-risk locations for speeding violations with improved accuracy and speed, facilitating optimized resource allocation and enhanced road safety. Future studies are recommended to integrate weather condition data to further improve prediction accuracy.