The traffic volume of a roadway segment is of significant importance for several public and private sections of the industry. This volume is represented by the Annual Average Daily Traffic (AADT). The AADT expresses the average number of vehicles that travel daily on this particular roadway section within a year. The traditional method of estimating AADT is examined along with new methods in order to improve the accuracy of the predictions.
The literature review conducted at the beginning of this study comprises the theoretical background to develop the research methodology. The study data are provided from 2002 to 2007 by the Ohio Department of Transportation (ODOT). The first type of data is obtained from traffic counters that perform continuously throughout a year. The second type of data is generated by portable counters that record traffic volumes for a short-period of time. The prediction of the AADT is based on the combination of both types of data using several mathematical methods and newly developed statistical approaches.
The determination of seasonal adjustment factors (SAF) is the first step of the AADT estimation. Seven SAFs and five approaches of estimating the AADT are examined for thirteen individual vehicle classes and groups of classes. The most effective SAFs are selected based on the mean absolute error (MAE) and the standard deviation (SD) of the predictions. Two analyses are conducted for each step of the study: the first is based on SAFs estimated from the sum of the two directional volumes of a roadway, and; the second on SAFs calculated for each direction of the traffic. The continuous counters are grouped together using eight different combinations of traditional grouping techniques and cluster analysis. The k-means algorithm, a non-hierarchical clustering method, is used to group the continuous counters based on their monthly SAFs. Furthermore, a statistical-based method for determining the optimal number of clusters was developed. The results are consistent over time and show a significant improvement in the accuracy of the AADT when clustering is used. Based on the performance, the applicability and the practicality of the examined methods, geographical classification and cluster analysis were selected to generate the final factor groupings.
The assignment of short-term counts to counter groups includes the investigation of three methods: the traditional method; discriminant analysis, and; a new approach based on statistical similarities of traffic and temporal characteristics between a short-period count and factor groups. In total, fifty six assignment models were developed and compared. The analysis based on directional SAFs is more effective than the total volume analysis by 15% to 40%. The final results indicate that the statistical approach developed in this study results in a MAE and SD improvement over the traditional method by 51.75% and 67.73% correspondingly.
In addition to the traditional method, regression and Bayesian negative binomial techniques are examined to predict AADT. In total twelve models are developed with a training data set and the results are compared using a validation data set. Parameters of significance include the HPMS roadway functional classification, population density, spatial location and the average daily traffic. The results show a full Bayesian negative binomial model with a coefficient offset was the most efficient model framework for all four seasons of the year. This model was able to describe between 87% and 92% of the variability within the data set.