"(...) Understanding the different aspects of the labour market is key for the well-founded service provision of
public employment services. In Austria, labour market monitoring is performed by the Austrian Public
Employment Service (AMS), which systematically collects and analyses labour market aspects. The purpose of
this thesis was to support the AMS’ efforts in labour market monitoring by providing a supervised machine
learning model to estimate the (re-)integration likelihood of unemployed people into the Austrian labour
market. The machine learning approach applied was random forest, a widely used decision tree-based
ensemble method introduced by Leo Breiman in 2001. At the center of the study was the concept of
sustainable employment , a concept describing an employment relationship which lasts for at least 63 days.
The data for model building was drawn from the AMS database and included administrative data (i.e. data
obtained in the context of client support) and insurance data. From this data, two different data sets were
created, viz. the main set with 94 variables (comprising additional historical client information) and the
production set with 71 variables. The two performance measures used for assessing the performance quality
of the models were the area under the ROC curve (AUC) and the geometric mean (GM). The outcome of this
study were four final models of considerable prediction quality. (...)" |