A complexity-aware adaptive hybrid AI–ML framework for redundancy-aware feature optimization in intrusion detection benchmarks

Abstract This study investigates whether a coordinated redundancy-reduction and resource-aware learning pipeline can improve predictive performance and computational efficiency for intrusion-detection benchmark datasets. The framework combines correlation-based feature pruning with a resource-aware learning-control layer. The pruning stage reduces redundant predictors before model training, while the adaptive layer uses the observed resource/complexity state to regulate learning behavior. The study evaluates the resulting pipeline against conventional and static feature-selection baselines. The reported aggregate results in the present manuscript are retained only as descriptive values; dataset-specific repeated-run statistics are not inferred where they were not recorded in the original study. Accordingly, the claims are restricted to the evaluated intrusion-detection benchmarks and computational conditions rather than generalized to all big-data domains.