Hourly Bridge Freeway Traffic Crash Prediction Using Log Count Data Poisson Time Series INGARCH Model and Its Comparisons With Regression-based Methods

Traffic crash prediction is vital for relevant agencies to take precautionary measures to minimize the economic and social losses from traffic accidents. Currently, the popularity of machine learning, deep learning, and traditional regression-based models in crash predictions eclipsed the use of count data time series models. Count data model has many intrinsic advantages over machine learning based methods in crash analysis. It is an extension of conventional time series regression by extending normal distribution to Poisson and Negative binomial. Meanwhile, covariate variables can get properly incorporated and their influence on dependent variable is well interpreted. This study attempts to compare and examine the performances of the count data time series model with the regression-based models in hourly crash prediction, utilizing traffic crash data from the Sutong Yangtze River Bridge in China. Log linear extension of Poisson distribution integer valued generalized autoregressive conditional heteroscedasticity models (INGARCH), as a type of count data model, is adopted and compared with the zero-inflated Poisson model (ZIP), as well as the cumulative link model for ordinal regression (CLM). The performances of ZIP and log linear extension of INGARCH count data model are similar and superior to the performances of CLM. Results showed that previous traffic accidents influence the crash occurrence in the near future and the employment of count data time series model in hourly crash prediction can appropriately capture this influence, with an average model sensitivity rate of 77.5%.

Count Data Time Series Modelling in Julia—The CountTimeSeries.jl Package and Applications

A new software package for the Julia language, CountTimeSeries.jl, is under review, which provides likelihood based methods for integer-valued time series. The package’s functionalities are showcased in a simulation study on finite sample properties of Maximum Likelihood (ML) estimation and three real-life data applications. First, the number of newly infected COVID-19 patients is predicted. Then, previous findings on the need for overdispersion and zero inflation are reviewed in an application on animal submissions in New Zealand. Further, information criteria are used for model selection to investigate patterns in corporate insolvencies in Rhineland-Palatinate. Theoretical background and implementation details are described, and complete code for all applications is provided online. The CountTimeSeries package is available at the general Julia package registry.