Intelligent Waste Management: A Multi-Modal AI Framework for Real-Time Waste Classification, Routing Optimization, and Circular Economy Integration

Abstract

Rapid urbanization and limited digital infrastructure challenge waste management in cities like Tripoli, Libya. This study proposes a simulation-based AI framework integrating multi-modal vision transformers for waste classification, reinforcement learning for dynamic routing, and a knowledge graph for circular economy matching. Using a synthetic environment built from regional audits, OpenStreetMap, and socio-spatial data, the framework reduced collection distance by 28–32%, improved classification F1-score to 84.7%, and increased recycler matching by 2.6× over fixed-schedule baselines across 50 Monte Carlo runs (p < 0.001). The work provides a policy-ready blueprint for data-scarce cities aligned with UNSustainable Development Goals (SDGs) 11 and 12.