Application of Outlier Detection and Missing Value Estimation Techniques to Various Forms of Traffic Count Data

This paper reports on the application of suitable techniques for detecting outliers and suggesting estimates for missing values in various forrns of traffic count data. The data used in this study came from three sources. The first set was provided by the Department of Transport's (DOT) regional office in Leeds and consists of automatic hourly traffic counts at four sites. The second set was part of a larger database provided by West Yorkshire Highways, Engineering and Technical Services (HETS). This set consists of automatic half hourly traffic counts on a single site. The third and final set was provided by Nottinghan University and consists of automatic five minute traffic counts at 40 locations, in close proximity to each other, from Leicester.

Three suitable techniques emerged from pilot studies of such series conducted by Watson et a1 (1992a) and Redfern et a1 (1992). The three techniques are: a) Maintaining an average and variability measure over time; b) ARIMA modelling with detection of large residuals; C) A point's influence on the correlation structure of the series. A fourth technique, by-eye detection and estimation, provides an intuitive comparison for the first three techniques.