2018 Schwemmle, Nicola: Modeling Automatic Visitor Count Data

Diese Abschlussarbeit war Teil des Projektes "Standardentwicklung Besuchermonitoring".
The following thesis analyzes visitor data that was automatically recorded from several sensors in the "Nationalpark Schwarzwald". The main part is devoted to error spotting and replacing. The presented approach to visual error spotting is designed to be applicable for all visitor data that will be recorded in the park in the future. Based on the assumption that there are very low visitor numbers during the night, nighttime outliers are identified and visually inspected. Out of 13 suspicious observations, three observations are identified to be most likely errors. In a second step, erroneous observations are replaced by fitting various linear models to the data and selecting the best fit with a test set. It turns out that, although having the lowest test error in one replacement process, the Poisson-model, which is specically designed for count data, does not always perform best. The data sets with replaced observations are then used to test assumptions about differences in seasonal patterns between the sensors. This analysis shows that the different locations of the sensors have a negligible impact on the seasonal patterns if aggregated observations are inspected.