Feature Normalization and Comparative Evaluation of Classification Techniques for Wireless Ad Hoc Networks

Abstract

 Wireless ad hoc networks (WANETs) are self-organizing systems without centralized infrastructure, which makes them highly dynamic but vulnerable to performance degradation due to feature variability and noise in network parameters. This paper proposes a framework for feature normalization and identification of suitable features for efficient classification and prediction in WANETs. Standardization methods such as Min-Max scaling and Z-score normalization are applied to key network features including node degree, mobility rate, packet delivery ratio, and energy consumption. A comparative analysis is performed between machine learning techniques (Random Forest, Support Vector Machine, and Deep Neural Networks) using normalized vs. unnormalized features. Experimental results on a simulated ad hoc network environment (NS-3) demonstrate that normalized features enhance model accuracy by up to 12% and reduce training variance, outperforming baseline models. The study establishes that feature normalization is crucial for robust and scalable WANET optimization.