Dos attack detection using fuzzy temporal deep long Short-Term memory algorithm in wireless sensor network
P. Sathishkumar, A. Gnanabaskaran, M. Saradha, R. Gopinath
Abstract
Wireless sensor networks (WSNs) are becoming more prevalent for data collection and monitoring. Compared to other sensing methods, each sensor node is typically tiny, inexpensive, and simple to implement. Consequently, a wide range of applications and industries utilize WSNs extensively. WSNs, on the other hand, are vulnerable to various security attacks and threats. Because they require more resources (including power, storage, processing power, and bandwidth) to develop defenses, it is necessary to have an efficient Intrusion Detection System (IDS) to detect these attacks and ensure security, even with these constraints. Traditional IDSs are losing effectiveness due to the increasing intelligence, frequency, and sophistication of malicious attacks . One of the most common attacks threatening WSNs is denial of service (DOS). To overcome these challenges, this research work proposes a novel fuzzy model incorporating the Deep Long Short-Term Memory (LSTM) algorithm for intrusion detection in WSNs. The proposed algorithm uses temporal constraints and fuzzy rules for weight fitting in decision-making, as well as neural networks in deep LSTM classifiers. The proposed method consists of (i) a pre-processing stage for preparing the data for further evaluation, (ii) a dynamic feature selection stage to select the most adaptable and efficient features and reduce processing time, and (iii) a detection stage to identify Denial-of-Service (DoS) attacks. It introduces the Crow Search Algorithm (CSA) for feature selection, aimed at optimizing feature efficiency. The Fuzzy Logic with LSTM model introduces a unique methodology where multiple sensor nodes collaboratively train a central global model while protecting private data and addressing privacy concerns effectively. This approach enables the model to detect sophisticated and previously unknown cyber threats by analyzing local and temporal correlations within network patterns, specifically tailored for identifying various types of DoS attacks. The model utilizes specialized KDDCup99 and NSL-KDD datasets. The suggested FL-LSTM approach achieved the highest accuracy (99.58%), the highest precision (98.42%), the highest recall (98.45%), and the highest f-score (98.36%). Compared to the conventional algorithm, the proposed FL-LSTM outperforms.