@inproceedings{NadimpalliBugertBittneretal., author = {Nadimpalli, Vamsi Krishna and Bugert, Flemming and Bittner, Dominik and Staufer, Susanne and R{\"o}hrl, Simon and Hauser, Florian and Ezer, Timur and Grabinger, Lisa and Maier, Robert and Mottok, J{\"u}rgen}, title = {Probabilistic Machine Learning for Simulating Complex Learner Profiles}, series = {2024 21st International Conference on Information Technology Based Higher Education and Training (ITHET), 06-08 November 2024, Paris, France}, booktitle = {2024 21st International Conference on Information Technology Based Higher Education and Training (ITHET), 06-08 November 2024, Paris, France}, publisher = {IEEE}, isbn = {979-8-3315-1663-5}, issn = {2473-2060}, doi = {10.1109/ITHET61869.2024.10837641}, pages = {1 -- 10}, abstract = {Data is the foundation of all machine learning applications. In education science, especially for the learner characteristics that drive personalized learning, it is difficult to collect and often uncertain. It is challenging to model, train, evaluate, and analyze the underlying algorithms when developing AI-based systems and having small sample sizes. To address these problems, we present a synthetic data generator utilising probabilistic models. This generator can effectively model and simulate complex learner profiles. To achieve this, we collected extensive data on learning styles, learning strategies, personalities, and preferred learning paths from 593 students over several semesters at a higher education level. Then, Bayesian networks, Hidden Markov Models, and Markov Chains are used to model the relationships between learner profiles. Using the Bayesian information criterion, and cross-validation with log-likelihood scores, we compare various models to select the best fitting one for synthesizing the data. The synthetic data is then evaluated using statistical validation techniques. In addition, we developed a simulation module with the option to simulate learner profiles based on manual user-defined inputs. The data and code used in this work are available as open source1110.5281/zenodo.13768136 [Titel anhand dieser DOI in Citavi-Projekt {\"u}bernehmen] contributing to open science and developers for customized simulated data. In the future, this data will refine the training, evaluation, analysis, and benchmarking of algorithms for personalized learning.}, language = {en} } @inproceedings{NadimpalliStauferEzeretal., author = {Nadimpalli, Vamsi Krishna and Staufer, Susanne and Ezer, Timur and Bugert, Flemming and Bittner, Dominik and Hauser, Florian and Grabinger, Lisa and R{\"o}hrl, Simon and Maier, Robert and Mottok, J{\"u}rgen}, title = {Predicting learner characteristics using machine learning}, series = {ICERI 2024 Proceedings, 17th annual International Conference of Education, Research and Innovation,11-13 November, 2024, Seville, Spain}, booktitle = {ICERI 2024 Proceedings, 17th annual International Conference of Education, Research and Innovation,11-13 November, 2024, Seville, Spain}, editor = {G{\´o}mez Chova, Luis and Gonz{\´a}lez Mart{\´i}nez, Chelo and Lees, Joanna}, publisher = {IATED}, doi = {10.21125/iceri.2024.1454}, pages = {6005 -- 6014}, abstract = {In education science research, data collection is challenging due to difficulty identifying students at the higher education level, privacy concerns, and varying levels of student engagement. Importantly, psychological questionnaires can be lengthy, leading to incomplete responses. We conducted repeated studies, and over time, the focus of the research adapted, introducing new materials and consequently leading to missing learner characteristics in some datasets. In this research, the issue of incomplete learner characteristics is addressed using data from three different studies: winter term 2022/2023 (n=297), summer term 2023 (n=274), and winter term 2023/2024 (n=25). These studies collected various learner characteristics, such as learning styles, personalities, learning strategies, and learning element preferences. However, learning element preferences and learning strategies were missing in the winter term of 2022, and the summer term of 2023 respectively. To analyze the data and predict these missing features, statistical analysis, and machine learning techniques were employed. Then, these models are rigorously evaluated using cross-validation and performance metrics like accuracy, precision, recall, and F1-score. Our findings provide insights into the relationships between learners' learning styles, personalities, learning strategies, and learning element preferences. This offers valuable implications for the design and implementation of educational interventions, like learning path recommendations. The results imply that machine learning models can predict missing learner characteristics, thus addressing the problem of incomplete data in educational research}, language = {en} } @inproceedings{StauferBugertHauseretal., author = {Staufer, Susanne and Bugert, Flemming and Hauser, Florian and Grabinger, Lisa and Ezer, Timur and Nadimpalli, Vamsi Krishna and Bittner, Dominik and R{\"o}hrl, Simon and Mottok, J{\"u}rgen}, title = {Tyche algorithm: Markov models for generating learning paths in learning management systems}, series = {INTED2024 Proceedings: 18th International Technology, Education and Development Conference, Valencia, Spain. 4-6 March, 2024}, booktitle = {INTED2024 Proceedings: 18th International Technology, Education and Development Conference, Valencia, Spain. 4-6 March, 2024}, editor = {G{\´o}mez Chova, Luis and Gonz{\´a}lez Mart{\´i}nez, Chelo and Lees, Joanna}, publisher = {IATED Academy}, isbn = {978-84-09-59215-9}, issn = {2340-1079}, doi = {10.21125/inted.2024.1080}, url = {http://nbn-resolving.de/urn:nbn:de:bvb:898-opus4-71148}, pages = {4195 -- 4205}, abstract = {In the intricate tapestry of the cosmos, where celestial threads weave stories of destiny, an enigmatic deity stands at the intersection of chance and fate — Tyche, the goddess of fortune. In science and probability, coincidence plays a distinctive role in Bayesian Networks (BNs) and Markov Models (MMs). This paper introduces the Tyche algorithm named after the goddess of fortune. The Tyche algorithm is a Markov model designed to generate learning paths in Learning Management Systems (LMSs). A learning path is a type of individualization that personalises the order of learning elements within an LMS course. Learning elements are fundamental components within an LMS course, depicting the learning content in diverse ways. In our case, the learning path tailoring is based on the learner's learning style according to Felder-Silverman Learning Style Model (FSLSM) - an indicator for the ideal pathway and learning element for the learner's optimum learning. The Tyche algorithm offers the advantage to provide students the most suitable learning path. Tyche is a MM structure with various matrices containing transition probabilities depending on the learning style. Nine categories of learning elements defined in a previous survey form the basis of the node structure of the MM. For a generic approach, a survey was designed to obtain the transition probabilities depending on the individual learning style. The survey with more than 100 German students participated is processed with the tool LimeSurvey. Students are asked about their learning style using the Index of Learning Styles (ILS) questionnaire according to Felder-Silverman and about the percentage probabilities of learning elements to get their individual sequence as learning path. The percentages are queried in two different ways. Firstly, the students were asked to provide a sequence of learning elements within each position of the sequence filled with probabilities for all learning elements. Secondly, a learning element is given and the students are asked to indicate which learning element they would work on next. The first way of asking for the probabilities is used to find the start node in the MM, whereas the second approach forms the probability matrices between the nodes within the MM. As result of the survey, the Tyche algorithm presents generic transition probabilities. It improves the learning process of individuals only by asking for their learning style: it generates individual learning paths through the learning elements within an LMS based on the MM explained above by solely getting the answers of the ILS questionnaire as input. In the future, other questionnaires such as BFI-10 for personality traits or LIST-K for learning strategies may offer a more comprehensive input. However, the next step is to evaluate Tyche with about 25 students in a software engineering lecture. This is planned for the year 2024.}, language = {en} } @misc{Bugert, author = {Bugert, Flemming}, title = {Applying Ariadne: Dataset on Learning Styles and Moodle-Based Learning Paths [Data set]}, doi = {10.5281/zenodo.12594911}, abstract = {This dataset contains information queried from 22 students inside a Moodle based learning management system during the winter term 2023/24 at a German university. Abstract With the use of learning management systems students benefit from being recommended suitable learning elements based on their individual needs. In doing so, recommendation algorithms are applied which first query the student's learning style. To improve the recommendation of learning elements a continuous analysis of the individual's learning style is required. A frequent questionnaire assessment would however be too time consuming. Instead, in a prior study an algorithm has been designed to identify changes in learning styles from the student's selection of learning elements. In this paper, we investigate the functionality of that algorithm by applying it on real student data. In particular, we test if the algorithm correctly indicates changes in learning styles. The utilised data is collected in our learning management system. To be precise, the data is obtained from 22 students enrolled in a software engineering course during the winter term of 2023/24. The data comprises two types of information for each student: 1) learning style collected at the start and end of the term, and 2) the user's actual selection of learning elements inside the learning management system. The uniqueness of this study lies in the data and the evaluation strategy based on it. Having the learning style at the end of the semester period as ground truth allows us to test if the algorithm operates correctly with actual user data from our learning management system. The results validate the behaviour of our algorithm, yet they strongly suggest the need for an adaptation. Further research is required on how to parameterise the underlying models.}, language = {en} }