A study on the construction of past travel time pattern for freeway travel time forecasting—focused on loop detectors—

The objective of this study is to propose methods for determining optimal representative value and the optimal size of historical data for reliable travel time forecasting. We selected values with the smallest mean of forecasting errors as the optimal representative value of travel time pattern data. The optimal size of historical data used was determined using the CVMSE (Cross Validated Mean Square Error) method. According to the results of applying the methods to point vehicle detection data of Korea Highway Corporation, the optimal representative values were analyzed to be the median. Because the mean used by previous researchers were influenced by each historical profile of travel time patterns, the distribution is easily skewed if there are very large or very small values in the travel time pattern data. Second, it was analyzed that 60 days' worth of data is the optimal size of historical data that should be used for travel time forecasting.