Adaptive Hammerstein- Wiener Model for Intelligent Networks

The rapid evolution of intelligent computing and next-generation networks necessitates models that accurately capture the intricate dynamics of time-dependent systems. This paper introduces an innovative adaptation of the Hammerstein-Wiener model for supervised learning tasks, specifically clas-sification and regression, within intelligent computing environ-ments. Traditional approaches to dynamic system modeling often struggle to balance linear and nonlinear dependencies, limiting their efficacy in domains demanding both predictive accuracy and adaptability. Our novel Hammerstein- Wiener model addresses this gap by combining nonlinear input transformations, a linear dynamic block to model temporal dependencies, and nonlinear output transformations. This unique configuration enables the model to capture complex, multi-step dependencies crucial for applications in network intrusion detection and other dynamic tasks. Evaluations on the KDD Cup 1999 dataset demonstrate the model's superiority over baseline methods, including Long Short-Term Memory (LSTM), achieving a classification accuracy of 98.97 %. Additionally, our model recorded a Prediction Stability Score (PSS) of 0.1693 and a perfect Time-weighted Accuracy (TWA) of 1.0, highlighting its remarkable temporal stability and computational efficiency. This adaptation of the Hammerstein-Wiener model establishes a robust, interpretable alternative for real-time dynamic systems within intelligent computing, offering a resource-efficient approach suitable for high-throughput applications in telecommunications, finance, and industrial au-tomation.