Previsão de horários dos ônibus do sistema de transporte público coletivo de Campina Grande.

Abstract

Predictability of public transport services is essential to improving its user experience. However,by working within a stochastic environment, predictability is typically impaired. In this work, we investigate the possibility of making a more predictable public transport system through the use of historical information, in a context where there is no available real-time vehicle location technology or updated information on the operation of the system. While GPS and other real- time Automatic Vehicle Location technologies (AVL) exists, many Brazilian cities do not have them available. Aware of this situation, we used data from the Campina Grande city bus system to evaluate the performance of four regression algorithms on the task of predicting, early in the day, how buses scheduled times will be fulfilled. Results show, although the lack of real time information may harm algorithms predictive ability in certain situations, using them makes it possible to forecast actual buses departure times with a median error of 28 seconds and buses arrival time with a median error of -167 seconds.