Abstract:
In this paper, we introduce FUSION (Feature Unification via Selection, Integration, and Optimization in Networks), an
innovative approach amalgamating various methods for optimal feature selection in wireless intruder detection systems. Incorporating
techniques based on filters, wrappers, embedded methods, and domain knowledge, FUSION is designed to effectively pinpoint
significant features in wireless networks, thereby enhancing the efficiency of intrusion detection. Our methodology initiates with a
comprehensive pre-processing stage. This stage focuses on normalizing and balancing the dataset, managing missing data, and
discarding irrelevant features. Beyond these pre-processing techniques, FUSION embraces a hybrid feature selection method,
harnessing the advantages of filter methods, suitable for initial feature screening, wrapper methods, proficient in interaction-based
selection, and embedded methods, which integrate feature selection within the model training process. A critical aspect of our
evaluation includes measuring the time taken for training for each feature selection method, providing insights into the computational
efficiency of the different techniques. To ensure the context's relevance throughout the selection process, we consider domain
knowledge. Decision-making within FUSION is influenced by a polling weight system, aggregating the selections made by different
classifiers, and prioritizing them accordingly. To verify the efficacy of our FUSION framework, we performed empirical evaluation.
The results underscored a significant enhancement in intrusion detection accuracy and provided a detailed analysis of the training time,
thus positioning FUSION as a promising approach to fortify network security within wireless systems.
