An Efficient Data Driven Model for WSN using Artificial Neural Network

Modern communication systems rely on WSNs as the most effective medium for remote monitoring and data acquisition across different environments. However, their inherent power constraints make providing security while maintaining efficiency in WSN nodes challenging. The existing solutions in the literature often fail to strike the best possible balance between security effectiveness and resource utilization. This paper proposes a data-driven intrusion detection system using a neural-network architecture optimized for a resource-constrained WSN environment. Based on the benchmark NSL-KDD dataset, our model uses medium-sized neural networks with activation functions and hidden layers to process patterns in the network traffic equally well. The study finds that our ANN-based approach outperforms the current techniques with 98.62% accuracy and a minimal computational cost of 2,053 units. This outcome surpassed all the comparable methods, including Gaussian SVM, which attained 98.52% accuracy with a cost of 2,194 units, and Decision Tree at a cost of 5,731 units, with an accuracy of 97.83% respectively.