Transferability analysis of user groups in travel behaviour surveys using a random forest classification model

Kurzfassung

There are many different travel behavior surveys with a wide range of sample sizes and contents, whose data sets are often difficult to merge. This contribution aims to evaluate the possibilities of transferring user typologies from one travel survey to another based on two different surveys in Berlin, Germany, using a random forest classification model. The investigated unimodal and intermodal mobility types were generated on the basis of a travel survey (n=1,098), collected in the year 2016 in Berlin with a special focus on intermodality, and were transferred into the Germany-wide survey “Mobility in Germany” (MiD) from 2017 (n=316,361 (total sample Germany); n=3,206 (subsample Berlin)). Basis for the training of the random forest model were mobility resources (e.g. public transport ticket availability, number of cars in household), socio-demographic characteristics (e.g. size of household, age, employment), temporally aggregated uni- and intermodal usage frequencies and trip purposes. At first, the model has been developed and tested based on the Berlin survey using different subsets of input variables (e.g. without usage frequencies, with usage frequencies, without intermodal usage) to evaluate which classification accuracy can be achieved depending on what kind of variables are included in the survey. In this process, gradual reduction of the variables was performed to evaluate the effects of using a reduced number of input variables for transfer. Based on usage frequencies, socio-demographics and mobility resources, the classification achieved a mean F1 score of 0.93 for the mobility types in the Berlin survey. Respectively, the results were lower when reducing the number of training variables. When performing the transfer, the distribution of the resulting user groups shows high similarities, especially for the Berlin sample. In conclusion, it can be shown that the proposed methodological procedure is suitable for transferring mobility types developed on the basis of a specialized data set to other surveys on the basis of mobility behavior parameters.