Non-GPS-based ETA Models Constructed from Historical GPS Data and Traffic Contexts

Without GPS data, ETA can be estimated with good precision from historical GPS data and the context of the traffic environment. In this paper, the estimated travel duration between bus stops was computed based on historical GPS data with time stamps. Three approaches were employed to compute the duration between bus stops: (i) statistical central tendencies, (ii) nonlinear regression techniques i.e., decision tree regressor, random forest regressor and k-nearest neighbors regressor, and (iii) an ANN predictive model. These three approaches capture different characteristics of the traffic environment. Statistical central tendencies give the overall view of the ETA information while nonlinear regressors can capture context from the time stamps (e.g., weekdays vs holidays). Finally, predicting the duration required to reach the next bus stop based on previous timespans on the same route implicitly captures temporal dynamics of the traffic environment from previous time steps and the ANN predictive model exploits this temporal information. The feasibility of ETA estimation using these three approaches was explained, evaluated and discussed.