Journal of Scientific Engineering Advances

Abstract

Efficient urban traffic management and transit HVAC optimisation depend heavily on systematic bus passenger monitoring, a task increasingly delegated to edge-deployed AI. Yet, numerous factors frequently influence the probabilistic confidence output in dynamic transit environments. This study strategically monitors, compares, and adapts the passenger monitoring via several key dimensions: regional supervision, resolution sensitivity scaling, data-efficient learning, inference calibration, and probabilistic confidence outputs in Artifiical intelligence. We investigate the latent bridges within the chosen AI network to understand how these coordinated setups enhance model robustness while systematically identifying and eliminating extraneous configurations. Our multi-vector optimization reveals that integrating uncertainty-aware confidence outputs with adaptive resolution sensitivity unleashes the potential of a deployment-oriented framework, advancing intelligent transportation and providing an adaptive framework for dependable crowd analytics.