Abstract—Flooding, blackhole, and forwarding are Wireless Sensor Networks (WSNs) attack types that have a severe impact on the reliability of the network and the integrity of data. This paper reports on a detailed benchmarking and hybridization analysis of seven deep learning models regarding multiclass intrusion detection on the WSNBFSF dataset. Comparison is made of the Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and their mixtures, CNN-RNN, CNN-LSTM, and RNNLSTM. The same experimental conditions were used to train all the models and make the evaluation fair. The experimental findings are indicative of the fact that the ANN has the highest performance with accuracy, precision, recall, and F1−Score of 99.81% in split validation and 99.89% in 5- fold cross-validation, offering close to perfect discrimination of all attack types. The LSTM (98.27%) and RNN-LSTM (97.78%) memory augmented models also demonstrate a high ability to capture traffic time dependencies in WSNs, whereas the CNN-based models are efficient in spatial pattern extraction but slightly worse on fine-grained attack detection. Conversely, a standalone RNN has problems of vanishing gradients and thus gives an accuracy of 84.77 percent. In summary, the results provided serve as a pointer to architecture-specific strengths, and they provide important information in designing a scalable, real-time, and resource-efficient deep learning-based intrusion detection system with the peculiarities of WSN environments.


Cite: Omar Almomani, Areen M. Arabiat, Al-Ahmed Hind, and Esraa Alsariera, “Hybridization of Deep Learning Models for Multiclass Attack Detection in Wireless Sensor Networks," Journal of Communications, vol. 21, no. 1, pp. 20-34, 2026.