Public transport congestion detection using incremental learning

Abstract

In the past decade, intelligent transportation systems have emerged as an efficient way of improving transportation services, while machine learning has been the key driver that created scopes for numerous innovations and improvements. Still, most machine learning approaches integrate paradigms that fell short of providing cost-effective and scalable solutions. This work employs long short-term memory to detect congestion by capturing the long-term temporal dependency for short-term public bus travel speed prediction to detect congestion. In contrast to existing methods, we implement our solution as incremental learning that is superior to traditional batch learning, enabling efficient and sustainable congestion detection. We examine the real-world efficacy of our prototype implementation in Pécs, the fifth largest city of Hungary, and observed that the incrementally updated model can detect congestion of up to 82.37%. Additionally, we find our solution to evolve sufficiently over time, implying diverse real-world practicability. The findings emerging from this work can serve as a basis for future improvements to develop better public transportation congestion detection.