Adapting and Employing Smart City Sensor Data for Strategic Planning

Thumbnail Image
Journal Title
Journal ISSN
Volume Title
Today, more and more cities are adopting the ‘smart city’ trend by deploying new smart devices, trackers and sensors in public spaces. They gather data about the city life dynamics in order to make city planning and maintenance more efficient and effective. However, currently, most smart systems are heavily data-driven and require vast amounts of information for training before they can deliver any useful insights. The focus of this study is one of the straight-forward applications for pedestrian traffic data – building a prediction model (by using pedestrian traffic data from the city of Nijmegen (the Netherlands)). The aim is to explore different prediction methods: multi-layer perceptron, Gaussian process and support vector regression models, compared to an averaging-based baseline model and find one that performs the best with only a year or less of training data. Then, attempt to improve the applicability of that model further with the conversion of single value prediction to a prediction range as well as applying spatial interpolation to gain insight about unobserved areas in the city. The results show that a simple averaging-based model performs the best, given a low complexity version of the problem (only 168 possible value combinations for the input variables), which highlights the importance of problem analysis, while a described attempt of radial basis function interpolation of spatially sparse observations (predictions), resulting in only very high-level insights, shows how impactful problem representation is for the results of the system
Faculteit der Sociale Wetenschappen