Descriptive and Predictive Modelling Techniques for Educational Technology

Data-driven models are the basis of all adaptive systems. Adaption to the user requires that the models are driven from real user data. However, in educational technology real data is seldom used, and all general-purpose learning environments are predefined by the system designers. In this thesis, we analyze how the existing knowledge discovery methods could be utilized in implementing adaptivity in learning environments. We begin by defining the domain-specific requirements and restrictions for data modelling. These properties constitute the basis for the analysis, and affect all phases of the modelling process from the selection of the modelling paradigm and data preprocessing to model validation. Based on our analysis, we formulate general principles for modelling educational data accurately. The main principle is the interaction between descriptive and predictive modelling. Predictive modelling determines the goals for descriptive modelling, and the results of descriptive modelling guide the predictive modelling. We evalute the appropriateness of existing dependency modelling, clustering and classification methods for educational technology, and give special instructions for their applications. Finally, we propose general principles for implementing adaptivity in learning environments. Computing Reviews (1998)