A Hybrid Deep Learning Model for Intrusion Detection System in the Internet of Things Environment
Vandana Choudhary, Sarvesh Tanwar, Tanupriya Choudhury · 2023
With billions of interconnected devices on the Internet, the Internet of Things (IoT) has emerged as a fundamental facet of our daily existence, influencing everything from our lifestyles to our work routines. This massive connectivity of devices to the Internet has put IoT systems at risk of cyber-attacks. The integration of IoT devices into various verticals has brought about unseen opportunities and challenges as well. There exists a need for robust as well as efficient Intrusion Detection Systems (IDSs) keeping in view the growing complexity of IoT networks. IDS plays a pivotal role in safeguarding the IoT environment by identifying malicious activities. In addition to traditional IDS approaches, contemporary approaches face challenges in effectively identifying the dynamic IoT network traffic as normal or malicious. To deal with this, we propose a novel hybrid deep learning classification model (CNN-LSTM), tailored specifically for IoT-based IDS. The proposed CNN-LSTM model capitalizes on the characteristics of both CNN and LSTM architectures. Our study employs a benchmark dataset, IoT-23 to train and evaluate the CNN-LSTM model. The dataset is first pre-processed, followed by feature selection by means of the Chi-Square Test. Finally, the Synthetic Minority Over-sampling Technique (SMOTE) is applied to balance the dataset before training the proposed model. The experimental findings showcase that our proposed model surpasses accuracies achieved by Logistic Regression, Naïve Bayes, CNN, LSTM models, and other existing models, achieving an impressive accuracy of 99.98%. By tapping into the power of deep learning, this model demonstrates outstanding potential in efficiently securing IoT networks from cyber-attacks leading to the construction of safe and reliable IoT ecosystems. In conclusion, the proposed model displays robustness against previously unseen intrusion patterns, illustrating its ability to respond to evolving cyber threats in the dynamic IoT world.