Smart deep learning model for enhanced IoT intrusion detection.

Area of Science:

• Cybersecurity
• Machine Learning
• Network Intrusion Detection

Background:

• The Internet of Things (IoT) generates vast amounts of data, necessitating robust cybersecurity to combat sophisticated cyberattacks.
• Machine learning-based anomaly detection offers a promising approach for identifying abnormal network traffic indicative of intrusions.
• Current methods struggle with evolving threats due to limitations in preprocessing and hyperparameter tuning of conventional models.

Purpose of the Study:

• To address limitations in existing intrusion detection systems by optimizing machine learning and deep learning models.
• To improve multiclass intrusion detection accuracy and efficiency across diverse datasets.
• To enhance the detection of complex, evolving cyber threats in IoT environments.

Main Methods:

• Implemented extensive preprocessing steps followed by Grid Search hyperparameter tuning for XGBoost and Sequential Neural Network (OSNN) algorithms.
• Augmented deep learning models with various filters, kernels, activation functions, and regularization techniques.
• Comprehensively tested the proposed system on NSL-KDD, UNSW-NB15, and CICIDS2017 datasets.

Main Results:

• Optimized XGBoost achieved 99.93% accuracy on NSL-KDD, with high F1-score and MCC, and low False Positive Rate (FPR).
• Optimized SNN demonstrated 99.0% accuracy and 1.00 AUC on NSL-KDD.
• The OSNN model achieved 96.80% accuracy on UNSW-NB15 and 99.53% accuracy on CICIDS2017, with low loss values.

Conclusions:

• The optimized OSNN model's superior performance is attributed to careful hyperparameter tuning, including activation functions, learning rates, and regularization.
• The proposed method demonstrates significant potential for enhancing intrusion detection, system integrity, and fraud prevention.
• This approach offers a pathway to optimize overall network performance in the face of advanced cyber threats.