Exploring the Applications of Machine Learning in the Public Sector

Abstract: Despite the many use cases for machine learning, it sees minimal usage in Sweden’s public sector today. It is important that the public sector in particular utilizes the most efficient tools available. For this reason, the Department of Employment and Social Services in Malmö has started a project to explore the viability of introducing machine learning to their work. This thesis corresponds to roughly the first half of the project. In the first stages, data pre-processing was required to get the data in a machine readable format. The data was then visualized using two different methods. Unsupervised learning was applied through the use of hierarchical clustering. Both visualization and clustering are heavily dominated by binary variables. An autoencoder was used in order to reduce noise and dimensionality, but the dependence on certain binary features remained. The last step is supervised learning in the form of a random forest classification model, with the goal to predict individuals’ time spent at and financial aid received from the Department. This model appears able to extract information beyond the binary variables.