A Comparative Study of Parametric Regression Models to Detect Breakpoint in Traffic Fundamental Diagram

TL;DR

This study compares two data-driven methods, logistic function and two-regime models, for detecting traffic flow breakpoints, finding logistic functions slightly more accurate and computationally efficient.

Contribution

It provides a comparative analysis of logistic and two-regime models for traffic breakpoint detection using real urban freeway data.

Findings

Logistic function outperforms two-regime model in prediction accuracy.

Logistic function requires less computational effort.

Two-regime model helps identify transitional traffic states.

Abstract

A speed threshold is a crucial parameter in breakdown and capacity distribution analysis as it defines the boundary between free-flow and congested regimes. However, literature on approaches to establishing the breakpoint value for detecting breakdown events is limited. Most of existing studies rely on the use of either visual observation or predefined thresholds. These approaches may not be reliable considering the variations associated with field data. Thus, this study compared the performance of two data-driven methods, that is, logistic function (LGF) and two-regime models, used to establish the breakpoint from traffic flow variables. The two models were calibrated using urban freeway traffic data. The models'performance results revealed that with less computation efforts, the LGF has slightly better prediction accuracy than the two-regime model. Although the two-regime model had relatively lower performance, it can be useful in identifying the transitional state.