Toivonen, Tapani, and Ilkka Jormanainen. “Evolution of Decision Tree Classifiers in Open Ended Educational Data Mining.” In Proceedings of the Seventh International Conference on Technological Ecosystems for Enhancing Multiculturality, pp. 290-296. 2019.
URL1
Educational Data Mining (EDM) aims to produce new knowledge from educational settings to support educators, learners and other stakeholders. EDM aims to facilitate the understanding of the educational context by utilizing different methods of statistics and machine learning. Like wise to the current trends in data mining, also EDM approaches have shifted from black box tools and algorithms to more open-ended tools and algorithms where the EDM end-users can adjust multiple parameters, view visualizations, and even adjust the predictive models. Multiple studies have shown that the EDM end-users benefit from the white box approaches and tools. We introduce the concept of Augmented Intelligence (AUI) method in EDM. AUI method is applied in an iterative process where a white box machine learning algorithm generates a predictive model, which is adjustable by the EDM end-user. The adjustable predictive model affects to the perception of the end-user and the adjusting affects to the output of the predictive model. When applied in cycles, the AUI method generates new knowledge from the educational context. To study AUI method, a potential EDM end-user generated multiple adjustable decision tree models and we observed the evolution of the models. The study indicates that, over time, the models generalize better and the AUI method helps to avoid the issue of overfitting. Moreover, the study indicates that the cyclic nature of the AUI method facilitates deeper knowledge generation from the dataset, if the context is known by the end-user.