| Highway agencies at all levels invest a significant amount of financial and human resources in maintaining a strong traffic counting program. The backbone of this program is the data collected at permanent Automated Traffic Recorder (ATR) sites. ATRs provide Permanent Traffic Counts (PTCs). PTCs are of importance in understanding geographic, temporal, and longterm traffic variations. In particular, PTC data are used to establish road pattern groups and develop expansion factors to extrapolate short-period traffic counts (SPTC) into estimated Annual Average Daily Traffic (AADT) for the majority of road segments without permanent traffic counters. Analyses of PTC data from a few highway agencies indicated that a significant portion of PTCs contain missing data. The American Association of State Highway and Transportation Officials (AASHTO) recommended practice is to use only those PTCs that meet the requirements of minimum data quantity in each month of the year. However, an examination of missing data patterns show that many of the PTCs' data are in continuous blocks and, thus, a significant percent of PTCs have to be excluded from expanding SPTCs to AADT.;The literature review indicates that the practice of handling missing PTC data varies from jurisdiction to jurisdiction. Each agency seems to have its own minimum data quantity requirement, imputation method, maximum imputation limit, and AADT estimation approach. No research has been conducted to investigate the sensitivity of AADT estimation accuracy based on partial or imputed PTC data. In this study, about 21 years of PTC data from the province of Alberta, Canada, are used. The observations from the data analysis show that the agency is imputing base data during the period from 1984 to 1993. Hence, the distribution of missing data patterns and the estimation accuracy of AADTs are carried out using 11 years of traffic data, from 1995 to 2005, because they are “pure base data” without “pollutants” from any imputing activities. This study shows that fairly accurate AADT estimates can be obtained based on partial or imputed PTC data and suggests that AASHTO standards may be inappropriately stringent. Four methods, the simple statistical technique, traditional factor method, improved factor method, and a non-parametric regression method, namely k-Nearest Neighbours (k-NN), were used to impute missing PTC data, and it was found that, in general, the k-NN method resulted in more accurate AADT estimates than the other factor methods. |
