Based on Deep Learning Intelligent Information Systems: Architecture, Knowledge Integration & Decision Optimization

Abstract

The exponential growth of multi-source digital data in domains such as healthcare and manufacturing has intensified the need for intelligent information systems (IIS) capable of accurate, interpretable, and resource-efficient decision-making. However, existing deep learning-based IIS often suffer from fragmented knowledge integration, limited decision optimization, and inadequate explainability, restricting their generalization and compliance in real-world environments. This study proposes a deep learning-based IIS that unifies three core components: a Knowledge Integration Module (KIM) for semantic alignment of structured and unstructured data through graph-based fusion; a Hybrid Decision-Optimization Engine (HDOE) combining reinforcement learning and constrained optimization for adaptive decision control; and an Explainable Representation Layer (ERL) providing feature-level attribution to enhance transparency and auditability. Empirical evaluations on two public datasets, industrial and medical, demonstrate significant performance gains over four baselines: accuracy = 92.1±0.4 %, F1 = 91.7±0.5 %, and latency reduction = 18.7 %. Interpretability scores improved by 0.9 points, while cross-domain accuracy degradation remained under 5 % with noise-induced accuracy loss below 2.5 %. These results confirm that the proposed IIS achieves statistically verified improvements in efficiency, interpretability, and robustness. The framework provides a reproducible and explainable foundation for deep learning-based decision systems applicable to data-intensive, compliance-sensitive domains such as healthcare, finance, and industrial optimization.