Bus Passenger Load Prediction: Challenges from an Industrial Experience.

In times of ongoing pandemic outbreak, public transportation systems organisation and operation have been significantly affected. Among others, the necessity to implement in-vehicle social distancing has fostered new requirements, such as the possibility to know in advance how many people will likely be on a public bus at a given stop. This is very relevant for both potential passengers waiting at a stop, and for decision makers of a transit company, willing to adapt the operational planning. Within the domain of data-driven Intelligent Transportation Systems (ITS), some research activities are being conducted towards Bus Passenger Load (BPL) predictions, with contrasting results. In this paper we report on an academic/industrial experience we conducted to predict BPL in a major Italian city, using real-world data. In particular, we describe the difficulties and challenges we had to face in the data processing and mining steps, due to multiple data sources, with noisy data. As a consequence, in this paper we highlight to the ITS community the need of more advanced techniques and approaches suitable to support the instantiation of a data analytic pipeline for BPL prediction.