A case study of checking national household travel survey data with machine learning

In recent years, machine learning techniques have been increasingly tested and applied to physically collected data to optimize the processes. In this paper, machine learning is used to check travel survey data of the German Mobility Panel (MOP). In the MOP, verified and raw data have been available for several decades, on which algorithms can learn the practices of human checking routines. By using machine learning, the algorithm is expected to learn the checking patterns from the past and thus support the data checking of new datasets. To this aim, several algorithms are applied and tested. The presented model framework supports the identification of blatant deficits in the reports at the individual and trip levels. The neural network (NN) shows the most promising results as it decreases the number of data samples checked. The checking effort can be reduced by 20.4 % at the individual trip level. This work shows that machine learning can support the data checking process in the MOP at various levels, thus leading to significant time reduction.