Multiple piecewise regression for trip generation models

Abstract

Trip generation modelling is not a standard task as the relationship between the input and output variables is not exactly known and is often nonlinear. Furthermore, the input variables may have different types especially at the disaggregate levels. Accordingly, there is no single modelling technique that can be used for all trip generation cases. In this paper, a multiple piecewise regression technique is proposed to suit most trip generation cases. Thresholds can be identified at the discrete values in the case of categorical inputs; while, no thresholds are considered in the case of binary inputs. In addition, the locations of thresholds in case of continuous inputs are obtained to minimize the summation of squared errors and achieving the logical constraints. Thus, a broken hyperplane is determined using quadratic programming. The proposed technique is illustrated using a numerical example and then validated using three well-known examples. The results show that the proposed technique can fit the given data more accurately than the previous techniques. In addition, the proposed technique leads to more explicit and visualized models for practitioners. These findings will encourage researchers to apply the proposed technique in further applications such as trip distribution and mode choice models.