Användning av multivarianta metoder för klassificering av individuell migration

The decision to migrate tends to be coupled with significant changes in the migrant's personal life. Consequently, the ability to forecast which individuals are about to migrate would offer companies an excellent opportunity to benefit from this fact in their prospecting and marketing. Targeting imminent migrants with well-timed and suitable offers would yield higher readership levels due to the more interested prospects. The aim of this thesis was to develop a method for classifying the adult Finnish population into one of the following categories: those who will migrate within the following six months, and those who will migrate at a later date. Utilizing the k-nearest neighbour algorithm, generalized linear models, support vector machines, optimally-pruned extreme learning machines, and their ensembles this thesis sought to analyze which approach offers the best classification result and whether there are significant differences regarding the practical implementations of these methods. Provided access to the full extent of the Population Information System database, this thesis was offered a premium opportunity to model the Finnish migration phenomenon through a large sample size and an extensive set of features. On the other hand, this created a demand for feature selection, which in this thesis was conducted based on a literature review. Moreover, the Delta Test was implemented as a method for pruning the selected shortlist from additional redundant features. The results revealed that there exists a subset of features that have a substantial impact on the classification accuracy. In addition, it was concluded that all of the implemented methods were capable of producing satisfactory results and that their performances were rather similar. However, considering that this thesis was able to highlight some very distinct practical differences regarding the methods, these findings provide valuable insights concerning future implementations