Authors: Aravind Chagantipati
Abstract: A persistent obstacle in deploying artificial intelligence within enterprise network defense systems is the tension between operational accuracy and human transparency. Conventional, rule-guided systems such as decision trees lack the scale and sophistication required to detect multi-stage, contemporary cyber threats. Conversely, complex neural networks (including LSTM and CNN designs) provide superior classification rates but act as opaque models, making validation difficult within strict enterprise compliance structures. This study presents a tiered, dual-layered architecture that integrates high-capacity deep learning classifiers with post-hoc explainability engines, effectively bridging the gap between classification performance and regulatory auditing requirements.
DOI: http://doi.org/10.5281/zenodo.21412923
