Automatic knowledge discovery from sparse and large-scale educational data : case Finland

The Finnish educational system has received a lot of attention during the 21st century. Especially, the outstanding results in the first three cycles of the Programme for International Student Assessment (PISA) have made Finland’s education system internationally famous, and its unique characteristics have been under active research by various, predominantly educational, scholars since then. However, despite the availability of real but often sparse big data sets that would allow more evidence-based decision making, existing research to date has mostly concentrated on using classical qualitative and (univariate) quantitative methods. This thesis discusses, in general terms, knowledge discovery from large and sparse educational data—particularly from PISA—through the utilization and further development of multivariate data mining techniques and, more specifically, the application of these methods in the context of the Finnish educational system. Therefore, its goals are twofold and interrelated: to advance knowledge discovery methods and algorithms for sparse educational data to gain more interpretable models and to utilize these approaches to learn from the data and improve understanding of educational phenomena. This article-style dissertation is composed of 10 publications. The first publication provides a general knowledge discovery framework for analyzing sparse educational data. The succeeding seven publications discuss and advance methods for the special characteristics and complexities of PISA data and their usage for the quantitative educational knowledge discovery process. The final two publications demonstrate how human advising and decision making in Finnish educational institutions and related to the management of a national educational system can be automated and improved by employing the introduced analysis framework and process. All this provides new insights about Finnish education, advances the overall automatic quantitative knowledge discovery process, increases institutional awareness, and could save costs on various levels of the whole educational system.