Spatio-temporal clustering in application

Abstract

The importance of machine learning methods in the data analysis of both academic research and industry applications has advanced rapidly in recent years. This thesis will investigate how a method of unsupervised machine learning known as clustering can be employed to analyse spatial and spatio-temporal data from different fields of application. Spatio-temporal data present a particular challenge. In spatial contexts, the notion of dependency among geographically close elements needs to be considered when analysing the geographic distance as well as other spatial components. The temporal dimension of the data makes traditional dissimilarity metrics unsuitable due to the sequential ordering of data points. For this reason, this thesis will present ways of overcoming the shortcomings in existing methodologies when applied to these data types. By doing so, it will contribute to the literature on clustering through innovative extensions, adaptations, and considerations. The flexibility of clustering will be demonstrated in three different application contexts in health, finance, and marketing. As such, this thesis will also contribute to the academic literature in these areas and offer valuable insights into applicable machine learning methodology for practitioners.