A successful special session on “Advances in Learning Analytics and Educational Data Mining” was held at ESANN 2015 conference on 23rd of April. The special session was organized by SmartLab, University of Genoa and Eindhoven University of Technology, and 50 to 60 people attended. The speakers presented very interesting cases of LA and EDM in workplace, higher education and schools.
The session started by Mehrnoosh Vahdat who gave an introductory presentation on the state of LA and EDM. She highlighted the addressed issues, similarities and differences of LA and EDM, as well as the well-known methods applied in these areas, applications, and future trends in research and practice. European projects in LA and EDM from research to implementation and community building were briefly explained. She introduced the LACE project and Evidence Hub to the audience and encouraged them to later access the Evidence Hub to find cases in various fields.
The ESANN 2015 Special Session collected research results from various groups, dealing with issues related to both theoretical and practical application of LA and EDM approaches in different domains. The presented topics cover a vast range of techniques, ranging from descriptive approaches to predictive methods.
An interesting application domain of descriptive LA and EDM methodologies is related to ITS: when dealing with programming, such learning supporting systems mostly rely on pre-coded feedback provision, such that their applicability is restricted to modeled tasks. This causes researchers to report large authoring times for preparing and classifying the learning material. Benjamin Paassen from Bielefeld University, who received the best student paper award of the conference, presented the suitability of Machine Learning (ML) techniques to automate this process and compare Java programs.
Another interesting approach, which exists between ML and Human Learning (HL) was presented by Luca Oneto from University of Genoa: in order to maximize knowledge when dealing with an unknown phenomenon, humans, as algorithms, should grasp what originated experimental evidence, rather than simply memorizing them. In the presented work, ML methods to assess the generalization ability of learning algorithms are applied to human learning, for example, by designing exploratory experiments towards studying Human Algorithmic Stability.
Another important domain of application for LA and EDM is related to enhancing learning in occupational training programs. By gathering empirical data on multiple factors that can affect learning for work, and by applying computational approaches in order to describe, identify, and understand preconditions of effective learning, LA and EDM can be implemented to effectively get insights towards supporting social capital. Virpi Kalakoski from Finnish Institute of Occupational Health – Brain and Technology propose to combine theory- and data- driven approaches towards maximizing the usefulness of LA and EDM in this domain.
Some of the most used descriptive analytics approaches belong to the domain of clustering methods, which have been shown to be effective in several heterogeneous domains, both in academia and in industry. Tommi Kärkkäinen from University of Jyväskylä, presented an efficient version of a robust clustering algorithm for sparse educational data that considers weights, allowing generalization and tuning of a sample with respect to the corresponding population. The algorithm was used to divide the Finnish student population of PISA 2012 (namely, the latest data from the Programme for International Student Assessment) into groups, according to their attitudes and perceptions towards mathematics, for which one third of the data is missing.
Predictive methods have a remarkable importance in the field of supporting education and learning. An interesting application of predictive methodologies is related to pupils, who do not finish their secondary education. Stephen Alstrup from University of Copenhagen presented a case in which ML can be used to predict high-school dropout, thus enabling early interventions. It considers pupils who finished at least their first six months of Danish high-school education with the goal of predicting dropout in the next three months.
Additionally, the importance of predicting the performance of students based on their behavior and their characteristics was presented by Minoru Nakayama from Tokyo Institute of Technology. In a blended learning course, in particular, a participant’s note taking activity reflects learning performance, and the possibility of predicting performance in final exams is examined using metrics of participant’s characteristics and features of the contents of notes taken during the course. According to the results of this prediction performance, features of note-taking activities are a significant data source to predict final exam marks.
For more information please refer to:
M. Vahdat, A. Ghio, L. Oneto, D. Anguita, M. Funk, M. Rauterberg, “Advances in learning analytics and educational data mining,” 23th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN), Bruges, Belgium, 22-24 Apr. 2015.
View Twitter conversations and metrics using [Topsy]