Daily Traffic Count Imputation for Bicycle and Pedestrian Traffic: Comparing Existing Methods with Machine Learning Approaches

Abstract

Monitoring nonmotorized traffic is becoming increasingly common practice at local and state departments of transportation. These travel activity data are necessary to monitor the system and track progress toward active transportation policy and program goals. A common problem is that permanent count site data are often missing, making those sites less useful. Being able to accurately estimate those missing data records functionally increases the amount of data available to use by themselves as metrics for monitoring traffic but also makes available more data for factoring short-term sites. Using nonmotorized traffic counts from several cities in Oregon, this research compared the ability of day-of-year (DOY) factors, a statistical model, and machine learning algorithms to accurately impute daily traffic records for annual traffic estimation. Based on exhaustive cross-validation experiments using data not missing at random scenarios, this research concluded that random forest and DOY factor approaches could be used to impute daily counts for nonmotorized traffic but each approach comes with tradeoffs. Though for many missing data scenarios random forest performed best, this method is complicated to estimate and apply. DOY factor-based methods are simpler to create and apply, and though more accurate in scenarios with significant amounts of missing data, they were less flexible given the need for data from neighboring count sites. Negative binomial regression was also found to work well in scenarios with moderate to low amounts of missing data. This work can inform nonmotorized traffic count programs needing vetted solutions for traffic data imputation.

Applying an interpretable machine learning framework to study mobility inequity in the recovery phase of COVID-19 pandemic

“…information at the census tract level due to privacy protection, and such a phenomena is worsened during COVID-19 pandemic. In order to impute the missing information, the holt-winters exponential smoothing ( Roll, 2021 ) is utilized to the aggregated data. If a census tract has some trips that miss O.D.…”

Traffic Mosaic: Traffic Data Imputation Method Inspired by Image Super Resolution

“…A considerable number of studies have been conducted on missing traffic data imputation. At the same time, many traffic predictions emphasize the processing of missing data ( 9 – 15 ). A classification by Li et al divides traffic imputation techniques into three categories: prediction, interpolation, and statistical learning ( 7 ).…”

Imputing Missing Data in Hourly Traffic Counts

“…Using non-motorized count data from several cities in Oregon, a study investigated several different imputation methods [10]. It was concluded that random forest performed the best but was difficult to apply.…”