Özet:

Abstract Wireless Sensor Networks (WSNs) are pervasive in various domains due to their capability to monitor and collect data from the environment. However, the open and distributed nature of WSNs makes them susceptible to security threats and attacks. Detecting and mitigating these attacks is vital to confirming the veracity and reliability of the collected data. In this study, we propose a novel CNN-LSTM hybrid network that makes use of the geographical and temporal information found in sensor data to identify attacks in WSNs. The proposed hybrid network combines the advantages of long short-term memory (LSTM) networks and convolutional neural networks (CNNs). CNNs are used to automatically extract key features from the sensor input and learn spatial representations. In order to identify the temporal dependencies and long-term patterns present in the consecutive sensor readings, the output of the final CNN layer is then fed into the LSTM layers. We conducted experiments utilizing a real-world WSN dataset encompassing a variety of typical actions and various types of attacks to assess the efficacy of our technique. The dataset was carefully curated and labeled to ensure its representativeness and diversity. Our CNN-LSTM hybrid network was evaluated against a number of baseline models, such as standalone CNN and LSTM networks and conventional machine learning techniques frequently employed for attack detection in WSNs. The experimental findings show that, in terms of accuracy, precision, recall, and F1-score, our proposed CNN-LSTM hybrid network outperforms the baseline models. The hybrid network’s ability to capture both spatial and temporal information allows it to better discern subtle attack patterns that might be missed by the standalone CNN or LSTM models. Furthermore, the model exhibits robustness and generalization, effectively detecting various attack scenarios while maintaining low false positive rates.

Anahtar Kelimeler: