Establishing Anomalies in IoT Networks using Deep Learning for Industrial Internet of Things Security
Vamsi Krishna Pentela, Perumalsamy Deepalakshmi · 2025
The Industrial Internet of Things (IIoT) is transforming industrial automation by incorporating smart sensors, edge computing, and cloud-based analytics to optimize their operations and help improve efficiency and productivity. The proliferation of IIoT networks also has opened the floodgates to a plethora of threats, including network intrusions, data breaches and malicious attacks. Existing security mechanisms either fail to keep up with evolving cyber threats (in the case of rule-based intrusion detection systems (IDS)) or can't scale up to keep up with the data flooding into IIoT environments (like signature-based anomaly detection models). In this study, we introduce a new deep learning-based anomaly detection framework targeted at IIoT security through the detection of abnormal network activity and the prevention of potential cyber-attacks. The framework employs a hybrid deep learning model where CNNs are used for feature extraction and LSTM networks for sequential anomaly detection. The model being presented in this paper is trained from public IIoT datasets such as UNSW-NB15 and Bot-IoT datasets that are used categories of cyber threats including Distributed Denial-of-Service (DDoS) attacks, malware, and unauthorized access. We applied autoencoders to test feature selection while testing PCA and autoencoders to pre-process our data to ensure that models are efficient without comprising accuracy archived in our previous work. Experimental results show that the proposed deep learning model surpasses traditional machine learning algorithms (e.g. Support Vector Machines (SVM) and Random Forest (RF)) in precision, recall, and F1-score. The model surpasses 98% accuracy in anomaly detection accuracy and hence is seen to be robust enough to discover cyber threats in IIoT surroundings. In addition, this research tests the scalability of the deep learning model by implementing it at a real-time IIoT backend testbed using edge computing devices. These results suggest that deep learning-based anomaly detection represents a state- of-the-art mechanism for IIoT security, enabling early threat detection, eliminating false positive rates, and maintaining low levels of latency in the system. The recent findings originated from a research project exploring the use of deep learning-based techniques to bolster IIoT security measures for industrial networks, highlighting the need for comprehensive solutions for more effective threat detection and prevention in industrial environments. In the future, we will focus on federated learning to improve the efficiency of distributed anomaly detection at lower computational overhead.