Attention-Enhanced Deep Learning With AI-Driven Decision Making For 5G Slice Classification And Resource Allocation In Environmental Monitoring Systems

Abstract

An intelligent framework is proposed to enhance slice classification and resource allocation efficiency within 5G-enabled environmental monitoring environments. By integrating attention mechanisms with a lightweight classifier and a rule-guided post-decision layer, the system ensures high adaptability under multi-modal inputs, including service type, QoS constraints, and real-time environmental parameters. Non-linear analysis modules contribute to refined predictions under fluctuating traffic and sensor conditions. Evaluation demonstrates strong accuracy across both slice and resource categories, achieving over 96.5% in all key metrics. Training–testing curves and confusion matrices validate learning consistency, while cross-dataset trials confirm generalizability. Compared to recent deep learning models, the framework shows superior precision and minimal divergence error. Major challenges addressed include input heterogeneity, dynamic policy shifts, and maintaining decision integrity under uncertainty. The overall design emphasizes scalability, minimal error propagation, and responsiveness, supporting intelligent 5G service adaptation in evolving environmental contexts.