Travel Time Reliability Reference Manual/Imputing Missing Speed Data
Unlike INRIX, the NPMRDS data set does not include non real-time travel times or speeds. If real-time measurements do not exist for a given TMC during a given time period, the data point is left blank. In some cases, it may be desirable to impute some of the missing data points based on surrounding real-time data.
The figure below gives an example of imputing a missing data point based on surrounding spatial and temporal data.
Using a sample NPMRDS data set, an imputation was conducted in which missing data points were imputed only if the surrounding spatial and temporal points were all populated with real-time data. The percentage of missing data points which fit this criteria was around 3 percent. More often than not, the missing data points are not anomalies, but are accompanied by other missing points. This makes sense, since missing real-time data is usually the result of a lack of probe vehicles. A lack of real-time data in one TMC segment will likely result in neighboring TMCs to also lack real-time data. Similarly, surrounding temporal points are likely to lack real-time observations when vehicle volumes are low.