Abstract:
Cybersecurity threats have become more developed and are increasing with the growing adoption of IoT devices. So it is necessary to develop the effective Intrusion Detection Systems (IDS). This research focuses on designing an IDS that uses both machine learning and deep learning techniques to enhance detection accuracy and also mitigate cyber threats in IoT environments. Various machine and deep learning models, including Decision Tree, XGBoost, SVM, CNN, DNN, and LSTM were tested on different feature sets and sampling rates to evaluate their performance.Among the models, the Decision Tree classifier demonstrated the highest accuracy of 97%, followed closely by XGBoost at 95%. Deep learning models, particularly CNN, DNN, and LSTM, performed significantly better when trained on larger datasets, reaching an accuracy of 96% at higher sampling rates. In contrast, SVM struggled with complex attack patterns and achieved a maximum accuracy of 89%. The study also explored hybrid approaches, where a combination of traditional and deep learning models improved detection performance. The Hybrid Traditional Model attained 94% accuracy, while the Hybrid Deep Learning Model reached 92% accuracy, indicating the potential of integrated techniques for improving IoT security. Beyond model performance, this research underscores the importance of dataset size, feature selection, and sampling strategies in optimizing IDS efficiency. The results suggest that incorporating real-time anomaly detection, adversarial training, and ensemble learning could further enhance security measures against emerging threats. Future research should focus on real-world deployment, improving model adaptability, and optimizing feature engineering techniques to strengthen IoT network defenses. This research contributes to the ongoing effort to develop more intelligent and resilient security solutions for IoT ecosystems.
