Search

Is there a pattern in how students solve Parsons puzzles? Is there a difference between the puzzle-solving strategies of C++ and Java students? We used Markov transition matrix to answer these questions. We analyzed the solutions of introductory programming students solving Parsons puzzles involving if-else statements and while loops in C++ and Java from fall 2016 to fall 2020. We present the results of our analysis qualitatively as heat maps and quantitatively using descriptive statistics. We found that most students solved the puzzles in the order in which lines appeared in the correct solution. Counter-intuitively, we found this pattern even in the solutions of the puzzles involving nested if-else statements, multiple while loops and nested while loops. Students who solved the puzzles with the fewest actions acted upon fewer lines out of order, i.e., not in the order in which they appear in the final solution. Whenever we found a statistically significant difference between C++ and Java solutions, C++ solutions involved fewer out-of-order and redundant actions than Java solutions. We discuss the implications of these results for the use of Parsons puzzles as a tool for teaching introductory programming. [For the complete proceedings, see ED630829.]

The use of Bayesian Knowledge Tracing (BKT) models in predicting student learning and mastery, especially in mathematics, is a well-established and proven approach in learning analytics. In this work, we report on our analysis examining the generalizability of BKT models across academic years attributed to "detector rot." We compare the generalizability of Knowledge Training (KT) models by comparing model performance in predicting student knowledge within the academic year and across academic years. Models were trained on data from two popular open-source curricula available through Open Educational Resources. We observed that the models generally were highly performant in predicting student learning within an academic year, whereas certain academic years were more generalizable than other academic years. We posit that the Knowledge Tracing models are relatively stable in terms of performance across academic years yet can still be susceptible to systemic changes and underlying learner behavior. As indicated by the evidence in this paper, we posit that learning platforms leveraging KT models need to be mindful of systemic changes or drastic changes in certain user demographics. [For the complete proceedings, see ED630829. Additional funding was provided by the U.S. Department of Education's Graduate Assistance in Areas of National Need (GAANN) program.]

The Indian Institute of Science is proud to host the fully in-person sixteenth iteration of the International Conference on Educational Data Mining (EDM) during July 11-14, 2023. EDM is the annual flagship conference of the International Educational Data Mining Society. The theme of this year's conference is "Educational data mining for amplifying human potential." Not all students or seekers of knowledge receive the education necessary to help them realize their full potential, be it due to a lack of resources or lack of access to high quality teaching. The dearth in high-quality educational content, teaching aids, and methodologies, and non-availability of objective feedback on how they could become better teachers, deprive our teachers from achieving their full potential. The administrators and policy makers lack tools for making optimal decisions such as optimal class sizes, class composition, and course sequencing. All these handicap the nations, particularly the economically emergent ones, who recognize the centrality of education for their growth. EDM-2023 has striven to focus on concepts, principles, and techniques mined from educational data for amplifying the potential of all the stakeholders in the education system. The spotlights of EDM-2023 include: (1) Five keynote talks by outstanding researchers of eminence; (2) A plenary Test of Time award talk and a Banquet talk; (3) Five tutorials (foundational as well as advanced); (4) Four thought provoking panels on contemporary themes; (5) Peer reviewed technical paper and poster presentations; (6) Doctoral students consortium; and (7) An enchanting cultural programme. [Individual papers are indexed in ERIC.]

Prompting students to generate constructed responses as they read provides a window into the processes and strategies that they use to make sense of complex text. In this study, Markov models examined the extent to which: (1) patterns of strategies; and (2) strategy combinations could be used to inform computational models of students' text comprehension. Random Walk models further revealed how consistency in strategy use over time was related to comprehension performance. High school (n = 257) and college students (n = 153) produced constructed responses at predetermined points while reading a scientific text. Each constructed response was scored for the presence of three common comprehension strategies (i.e., paraphrasing, bridging, elaborating), such that each constructed response could then be categorized as one of eight combination types. Markov chains revealed that more and less successful comprehenders leveraged different comprehension strategies, such that skilled comprehenders were more likely to use combinations of strategies while reading the text, particularly paraphrasing and making connections between ideas within the text (i.e., bridging). Random Walk analysis further demonstrated that successful comprehenders employed strategies more consistently. The results demonstrate the utility of Markov and Random Walk models in profiling learners' strategy use based on their constructed responses. [For the full proceedings, see ED623995.]

This paper studies the use of Reinforcement Learning (RL) policies for optimizing the sequencing of online learning materials to students. Our approach provides an end to end pipeline for automatically deriving and evaluating robust representations of students' interactions and policies for content sequencing in online educational settings. We conduct the training and evaluation offline based on a publicly available dataset of diverse student online activities used by tens of thousands of students. We study the influence of the state representations on the performance of the obtained policy and its robustness towards perturbations on the environment dynamics induced by stronger and weaker learners. We show that 'bigger may not be better', in that increasing the complexity of the state space does not necessarily lead to better performance, as measured by expected future reward. We describe two methods for offline evaluation of the policy based on importance sampling and Monte Carlo policy evaluation. This work is a first step towards optimizing representations when designing policies for sequencing educational content that can be used in the real world. [For the full proceedings, see ED623995.]

A learner model reflects learning patterns and characteristics of a learner. A learner model with learning history and its effectiveness plays a significant role in supporting a learner's understanding of their strengths and weaknesses of their way of learning in order to make proper adjustments for improvement. Nowadays, learners have been engaging in online learning frequently and intensely, leaving behind tremendous learning history data that contain informative insights about the learners' learning patterns. This paper proposed a method for developing learner models by applying the Markov chain to learning history data. Our method transforms individual learners' resource use data in a learning course into a large amount of resource use sequences, then develops a Markov learner model, and generates the resource use steady state for each learner. The resource use density, the resource steady state, and the assessment scores of individual learners tell their learning patterns and the effectiveness of the learning patterns. From the Markov learner model, we generate a learner profile for describing learning patterns and an index for measuring the strength of learning habits of the learner. We verified our method by applying it to each course in the OULAD dataset to predict the learning performance using the index. The preliminary results gain up to 97% accuracy on the pass/fail prediction problem.

The teaching quality evaluation of college teachers is an important indicator of school evaluation, performance assessment and teacher promotion. However, the existing evaluation methods are only limited to the formal evaluation of teaching behavior, and have not achieved objective, reasonable and scientific evaluation results. The purpose of this paper is to propose a [delta]S equation of the state of the teaching system from the point of view of the teaching system's operation process. On this basis, a Markov process evaluation method for teachers' teaching quality is established. The research results show that the application of the [delta]S equation of the teaching system to the evaluation of teaching behavior and teaching effect has its own characteristics, and the key lies in the reasonable selection of the [delta]S equation and the the combination of Markov analysis can effectively achieve the evaluation goal of teachers' teaching quality. [For the full proceedings, see ED638044.]

In this article, the beta-binomial model for count data is proposed and demonstrated in terms of its application in the context of oral reading fluency (ORF) assessment, where the number of words read correctly (WRC) is of interest. Existing studies adopted the binomial model for count data in similar assessment scenarios. The beta-binomial model, however, takes into account extra variability in count data that has been neglected by the binomial model. Therefore, it accommodates potential overdispersion in count data compared to the binomial model. To estimate model-based ORF scores, WRC and response times were jointly modeled. The full Bayesian Markov Chain Monte Carlo (MCMC) method was adopted for model parameter estimation. A simulation study showed adequate parameter recovery of the beta-binomial model and evaluated the performance of model fit indices in selecting the true data-generating models. Further, an empirical analysis illustrated the application of the proposed model using a dataset from a computerized ORF assessment. The obtained findings were consistent with the simulation study and demonstrated the utility of adopting the beta-binomial model for count-type item responses from assessment data. [This paper was published in: The Texas Universities Educational Statistics and Psychometrics (TUESAP) Annual Meeting Proceedings, March 2023, pp. 1-26.]

This study examines data from a field experiment investigating the effects of a personalized recommendation algorithm that proposes to students which videos to watch next, after they complete mini-assessments for algebra that available on the Math Nation intelligent virtual learning environment (IVLE). The end users of Math Nation are students enrolled in an Algebra 1 course in middle and high schools of the state of Florida, and the IVLE is used both during and out of school time. The objective of the developed recommendation algorithm is to increase student preparation to take the state-mandated End-of-Course (EoC) Algebra 1 assessment at the end of the school year. The algorithm is based on a Markov Decision Process framework that uses as input the students' responses to a series of mini-assessment tests. The current study randomly assigned 16,406 students to either treatment or control conditions, which were blind to both students and teachers. The results indicate that the effects of the recommendation algorithm depend on the level of usage of students, showing significant improvements on EoC test scores of students who have a moderate level of usage. However, there was no effect for low usage students. The study also shows that students practicing with the mini-assessments available on Math Nation, helps them improve by a small margin their performance on the End-of-Course test, irrespective of the usage level. Finally, the study provides insights on challenges posed for implementing personalized recommendation algorithms at a large scale, related both to student self-regulation and teacher orchestration of technology use in the classroom. [For the full proceedings, see ED615472.]

With the emergence of MOOCs, it becomes crucial to automate the process of a course design to accommodate the diverse learning demands of students. Modeling the relationships among educational topics is a fundamental first step for automating curriculum planning and course design. In this paper, we introduce "Topic Transition Map" (TTM), a general structure that models the content of MOOCs at the topic level. TTMs capture the various ways instructors organize topics in their courses by modeling the transitions between topics. We investigate and analyze four different methods that can be exploited to learn the Topic Transition Map: (1) Pairwise Constrained K-Means, (2) Mixture of Unigram Language Model, (3) Hidden Markov Mixture Model, and (4) Structural Topic Model. To evaluated the effectiveness of these methods, we qualitatively compare the topic transition maps generated by each model and investigate how the Topic Transition Map can be used in three sequencing tasks: (1) determining the correct sequence, (2) predicting the next lecture, and (3) predicting the sequence of lectures. Our evaluation revealed that PCK-Means has the highest performance in the first task, HMMULM outperforms other methods in task 2, while there is no winning in task 3. [For the full proceedings, see ED615472.]

Bayesian Knowledge Tracing, a model used for cognitive mastery estimation, has been a hallmark of adaptive learning research and an integral component of deployed intelligent tutoring systems (ITS). In this paper, we provide a brief history of knowledge tracing model research and introduce pyBKT, an accessible and computationally efficient library of model extensions from the literature. The library provides data generation, fitting, prediction, and cross-validation routines, as well as a simple to use data helper interface to ingest typical tutor log dataset formats. We evaluate the runtime with various dataset sizes and compare to past implementations. Additionally, we conduct sanity checks of the model using experiments with simulated data to evaluate the accuracy of its EM parameter learning and use real-world data to validate its predictions, comparing pyBKT's supported model variants with results from the papers in which they were originally introduced. The library is open source and open license for the purpose of making knowledge tracing more accessible to communities of research and practice and to facilitate progress in the field through easier replication of past approaches. [For the full proceedings, see ED615472.]

Knowledge tracing algorithms such as Bayesian Knowledge Tracing (BKT) can provide students and teachers with helpful information about their progress towards learning objectives. Despite the popularity of BKT in the research community, the algorithm is not widely adopted in educational practice. This may be due to skepticism from users and uncertainty over how to explain BKT to them to foster trust. We conducted a pre-registered 2x2 survey experiment (n=170) to investigate attitudes towards BKT and how they are affected by verbal and visual explanations of the algorithm. We find that ostensible learners prefer BKT over a simpler algorithm, rating BKT as more trustworthy, accurate, and sophisticated. Providing verbal and visual explanations of BKT improved confidence in the learning application, trust in BKT and its perceived accuracy. Findings suggest that people's acceptance of BKT may be higher than anticipated, especially when explanations are provided. [For the full proceedings, see ED615472.]

One of the most challenging issues for online courseware engineering is to maintain the quality of instructional components, such as written text, video, and assessments. Learning engineers would like to know how individual instructional components contributed to students' learning. However, it is a hard task because it requires significant expertise in learning science, learning technology, and subject matter pedagogy. To address this challenge, we propose an innovative application of reinforcement learning (RL) as an assessor of instructional components implemented in given online courseware. After students activities are converted into Markov decision process (MDP), a collection of actions (each corresponds to an instructional component) suggested as a policy is analyzed. As a consequence, the usefulness of individual actions with regards to achieving ideal learning outcomes will be suggested. The proposed RL application is invented for human-in-the-loop learning engineering method called RAFINE. In the RAFINE framework, a machine generates a list of the least contributing instructional components on the given online courseware by interpreting the whole policy. The courseware developers modify those suggested components. As a proof of concept, this paper describes an evaluation study where online learning was simulated on hypothetical online courseware. The results showed that over 90% of ineffective instructional components were correctly identified as ineffective on average. [For the full proceedings, see ED607784.]

Massive online Open Courses (MOOCs) make extensive use of videos. Students interact with them by pausing, seeking forward or backward, replaying segments, etc. We can reasonably assume that students have different patterns of video interactions, but it remains hard to compare student video interactions. Some methods were developed, such as Markov Chain and Edit Distance. However, these methods have caveats as we show with prototypical examples. This paper proposes a new methodology of comparing video sequences of interaction based both on time spent in each state and the succession of states by computing the distance between the transition matrices of the video interaction sequences. Results show the proposed methodology can better characterize video interaction in a task to discriminate which student is interacting with a video, or which video a student is interacting with. [For the full proceedings, see ED607784.]

There has been considerable interest in techniques for modelling student learning across practice problems to drive real-time adaptive learning, with particular focus on variants of the classic Bayesian Knowledge Tracing (BKT) model proposed by Corbett & Anderson, 1995. Over time researches have proposed many variants of BKT with differentiation based on their treatment of the underlying parameters: (a) general across student and questions; (b) individualized for students; and (c) individualized for questions. Yet at the same time, most of these variants are similar in that they utilize the same Hidden Markov (HMM) architecture to model student learning and share many of the same drawbacks, including less effective balancing between recent and historical student data and assuming that students learn at the same rate across all the attempts irrespective of if they get the question right. At the same time, these variants share the virtue of parameter interpretability, a virtue not seen in recent efforts to recast knowledge tracing as a deep learning problem. This paper proposes a different architecture that replaces learning rate with recency weights which capture student improvement wholly through data rather than assuming constant learning across attempts and manages recent and historical data more appropriately while retaining the interpretability of BKT parameters. The proposed model was tested on multiple public datasets from ASSISTments and Mindspark and performed similarly to classic BKT model on unseen data. [For the full proceedings, see ED607784.]

Course selection is a crucial and challenging problem that students have to face while navigating through an undergraduate degree program. The decisions they make shape their future in ways that they cannot conceive in advance. Available departmental sample degree plans are not personalized for each student, and personal discussion time with an academic advisor is usually limited. Data-driven methods supporting decision making have gained importance to empower student choices and scale advice to large cohorts. We propose "Scholars Walk," a random-walk-based approach that captures the sequential relationships between the different courses. Based on the "wisdom of the crowd" and the students' prior courses, we recommend a short list of courses for next semester. Our experimental evaluation illustrates that Scholars Walk outperforms other collaborative filtering and popularity-based approaches. At the same time, our framework is very efficient, easily interpretable, while also being able to take into consideration important aspects of the educational domain. [For the full proceedings, see ED599096.]

Simplified categorizations have often led to college students being labeled as full-time or part-time students. However, at many universities student enrollment patterns can be much more complicated, as it is not uncommon for students to alternate between full-time and part-time enrollment each semester based on finances, scheduling, or family needs. While prior research has established that full-time students maintain better outcomes than their part-time counterparts, little study has examined the impact of mixed enrollment patterns on academic outcomes. In this paper, we apply a Hidden Markov Model to identify students' enrollment strategies according to three different categories: part-time, full-time, and mixed enrollment. According to the enrollment classification we investigate and compare the academic performance outcomes of each group. Analysis of data collected from the University of Central Florida from 2008 to 2017 indicates that mixed enrollment students are closer in performance to full-time students, than part-time students. More importantly, during their part-time semesters, mixed-enrollment students significantly outperform part-time students. Such a finding suggests that increased engagement through the occasional full-time enrollment leads to better overall outcomes. [For the full proceedings, see ED599096.]

The 12th iteration of the International Conference on Educational Data Mining (EDM 2019) is organized under the auspices of the International Educational Data Mining Society in Montreal, Canada. The theme of this year's conference is EDM in Open-Ended Domains. As EDM has matured it has increasingly been applied to open-ended and ill-defined tasks such as writing, design, and collaborative problem solving, and it has been used in new informal contexts where student actions are at best semi-structured. This iteration of the conference includes a range of work in these and other areas. This year's conference features three invited talks: Julita Vassileva, Professor at the Department of Computer Science, University of Saskatchewan, Canada; Steve Ritter, Co-Founder and Chief Scientist, Carnegie Learning Inc., Pittsburgh; and Michael Mozer, Professor Department of Computer Science and Institute of Cognitive Science University of Colorado. The number of accepted papers include 22 full papers and 42 short papers. An additional 47 papers were accepted to the poster track. The poster and demo track itself accepted 14 contributions out of 34 submissions. Together with the "Journal of Educational Data Mining" ("JEDM"), the EDM 2019 conference held a "JEDM" Track that provides researchers a venue to deliver more substantial mature work than is possible in a conference proceeding and to present their work to a live audience. The papers submitted to this track followed the "JEDM" peer review process. Two such papers are featured in the conference's program. Additionally this year, papers that were regularly published in the journal in 2018 were invited for presentation at the conference. Two authors accepted this invitation. The main conference invited contributions to an Industry Track in addition to the main track. The EDM 2019 Industry Track received eleven submissions of which six were accepted. The EDM conference continues its tradition of providing opportunities for young researchers to present their work and receive feedback from their peers and senior researchers. The doctoral consortium this year features eight such presentations. This year's conference includes also an invited talk by the authors of the 2018 winner of the EDM Test of Time Award. This year's talk is delivered by Mykola Pechenizkiy. In addition to the main program, there are three workshops: (1) Learning Analytics: Building bridges between the Education and the Computing communities; (2) Reinforcement Learning for Educational Data Mining; and (3) Workshop on EDM & Games: Leveling Up Engaged Learning with Data-Rich Analytics. Three tutorials were presented as well: (1) Sharing and Reusing Data and Analytic Methods with LearnSphere; (2) Causal Discovery with Tetrad in LearnSphere's Tigris and Designing and Developing Open; and (3) Pedagogically-Based Predictive Models using the Moodle Analytics API.

Simulations that combine real world components with interactive digital media provide a rich setting for students with the potential to assist knowledge building and understanding of complex physical processes. This paper addresses the problem of modeling the effects of multiple students' simultaneous interactions on the complex and exploratory environments such simulations provide. We work towards assisting educators with the difficult task of interpreting student exploration. We represent the system dynamics that result from student actions with a complex time series and use switch based models to decompose the time series into individual periods that target interpretability for teachers. The model learns the transition points between successive periods in the time series as well as the internal dynamics that govern each period. This model differs from other switch based models in that it decomposes the time series in a way that is human interpretable. This approach was applied to data that was obtained from an existing multi-person simulation with pedagogical goals of teaching sustainability and systems thinking. A visualization of the model was designed to validate the interpretability of the generated text-based descriptions when compared to a movie representation of the system dynamics. A pilot study using this visualization indicates that the switch based model finds relevant boundaries between salient periods of student work. [For the full proceedings, see ED593090.]

Knowledge tracing is a popular and successful approach to modeling student learning. In this paper we investigate whether the addition of neuroimaging observations to a knowledge tracing model enables accurate prediction of memory performance in held-out data. We propose a Hidden Markov Model of memory acquisition related to Bayesian Knowledge Tracing and show how continuous functional magnetic resonance imaging (fMRI) signals can be incorporated as observations related to latent knowledge states. We then show, using data collected from a simple second-language learning experiment, that fMRI data acquired during a learning session can be used to improve predictions about student memory at test. The fitted models can also potentially give new insight into the neural mechanisms that contribute to learning and memory. [For the full proceedings, see ED593090.]

In this paper, we present a system for visualizing learning logs of a course in progress together with predictions of learning activities of the following week and the final grades of students by state transition graphs. Data are collected from 236 students attending the course in progress and from 209 students attending the past course for prediction. From these data, the system constructs a state transition graph, where the prediction is based on the Markov property. We verify the performance of predictions by experiments in which the accuracy of prediction using the data of the course in progress and the one by 5-fold cross validation. [The research results have been achieved by "Research and Development on Fundamental and Utilization Technologies for Social Big Data" (178A03), the Commissioned Research of National Institute of Information and Communications Technology (NICT), Japan. For the complete proceedings, see ED579395.]

In this study, we applied decision trees (DT) to extract a compact set of pedagogical decision-making rules from an original "full" set of 3,702 Reinforcement Learning (RL)- induced rules, referred to as the DT-RL rules and Full-RL rules respectively. We then evaluated the effectiveness of the two rule sets against a baseline Random condition in which the tutor made random yet reasonable decisions. We explored two types of trees (weighted and unweighted) as well as two pruning strategies (pre- and post-pruning). We found that post-pruned weighted trees produced the best results with 529 DT-RL rules. The empirical evaluation was conducted in a classroom study using an existing Intelligent Tutoring System (ITS) named Pyrenees. 153 students were randomly assigned to three conditions. The procedure was the same for all students with domain content and required steps strictly controlled. The only substantive differences between the three conditions were the policy: (Full-RL vs. DT-RL vs. Random). Our result showed that as expected the machine induced policies (Full-RL and DT-RL) are significantly more effective than the random policy; more importantly, no significant difference was found between the Full-RL and DT-RL policies though the number of DT-RL rules is less than 15% of the number of the Full-RL rules and the former group also took significantly less time than the latter. [For the full proceedings, see ED596512.]

In this paper, we investigate two purported problems with Bayesian Knowledge Tracing (BKT), a popular statistical model of student learning: "identifiability" and "semantic model degeneracy." In 2007, Beck and Chang stated that BKT is susceptible to an "identifiability problem"--various models with different parameters can give rise to the same predictions about student performance. We show that the problem they pointed out was not an identifiability problem, and using an existing result from the identifiability of hidden Markov models, we show that under mild conditions on the parameters, BKT is actually identifiable. In the second part of the paper, we discuss a problem that has been conflated with identifiability, but which actually does arise when fitting BKT models, "semantic model degeneracy"--the model parameters that best fit the data are inconsistent with the conceptual assumptions underlying BKT. We give some intuition for why semantic model degeneracy may arise by showing that BKT models fit to data generated from alternative models of student learning can have semantically degenerate parameters. Finally, we discuss the potential implications of these insights. [For the full proceedings, see ED596512. For a related grantee submission, see ED577166.]

The analysis of log data generated by online educational systems is an important task for improving the systems, and furthering our knowledge of how students learn. This paper uses previously unseen log data from Edulab, the largest provider of digital learning for mathematics in Denmark, to analyse the sessions of its users, where 1.08 million student sessions are extracted from a subset of their data. We propose to model students as a distribution of different underlying student behaviours, where the sequence of actions from each session belongs to an underlying student behaviour. We model student behaviour as Markov chains, such that a student is modelled as a distribution of Markov chains, which are estimated using a modified k-means clustering algorithm. The resulting Markov chains are readily interpretable, and in a qualitative analysis around 125,000 student sessions are identified as exhibiting unproductive student behaviour. Based on our results this student representation is promising, especially for educational systems offering many different learning usages, and offers an alternative to common approaches like modelling student behaviour as a single Markov chain often done in the literature. [For the full proceedings, see ED596512.]

The extraction of student behavior is an important task in educational data mining. A common approach to detect similar behavior patterns is to cluster sequential data. Standard approaches identify clusters at each time step separately and typically show low performance for data that inherently suffer from noise, resulting in temporally inconsistent clusters. We propose an evolutionary clustering pipeline that can be applied to learning data, aiming at improving cluster stability over multiple training sessions in the presence of noise. Our model selection is designed such that relevant cluster evolution effects can be captured. The pipeline can be used as a black box for any intelligent tutoring system (ITS). We show that our method outperforms previous work regarding clustering performance and stability on synthetic data. Using log data from two ITS, we demonstrate that the proposed pipeline is able to detect interesting student behavior and properties of learning environments. [For the full proceedings, see ED592609.]

Research has shown that supporting tinkering and exploration promotes a wide range of STEM related literacies. However, the open-endedness of tinkering environments makes it difficult to know whether learners' exploration is productive or not. This is especially true in museum spaces, where dwell times are short and facilitators lack a history of engagement with individual visitors. In response, this study uses telemetry data from Oztoc -- an open-ended exploratory tabletop exhibit in which visitors embody the roles of engineers who are tasked with attracting and cataloging newly discovered aquatic creatures by building working electronic circuits. This data is used to build Hidden Markov Models (HMMs) to devise an automated scheme of identifying when a visitor is behaving productively or unproductively. Evaluation of our HMM was shown to effectively discern when visitors were productively and unproductively engaging with the exhibit. Using a Markov model, we identify common patterns of visitor movement from unproductive to productive states to shed light on how visitors struggle and the moves they made to overcome these struggles. These findings offer considerable promise for understanding how learners productively and unproductively persevere in open-ended exploratory environments and the potential for developing real time supports to help facilitators know how and when to best engage with visitors.

Large-scale experiments are often expensive and time consuming. Although Massive Online Open Courses (MOOCs) provide a solid and consistent framework for learning analytics, MOOC practitioners are still reluctant to risk resources in experiments. In this study, we suggest a methodology for simulating MOOC students, which allow estimation of distributions, before implementing a large-scale experiment. To this end, we employ generative models to draw independent samples of artificial students in Monte Carlo simulations. We use SemiMarkov Chains for modeling student's activities and ExpectationMaximization algorithm for fitting the model. From the fitted model, we generate simulated students whose processes of weekly activities are similar to these of the real students. [For the full proceedings, see ED592609.]

Online programming discussion forums have grown increasingly and have formed sizable repositories of problem solving-solutions. In this paper, we investigate programming learners' information seeking behaviors from online discussion forums. We design engines to collect students' information seeking processes, including query formulation, refinement, results examination, and reading processes. We model these behaviors and conduct sequence pattern mining. The results show that programming learners indeed seek for programming related information from discussion forums by actively searching on the site and reading posts progressively according to course schedule topics. Advanced students consistently perform query refinements, examine search results and commit to read, however, novices do not. In addition, advanced students commit to read posts, but novices only skim. [For the full proceedings, see ED592609.]

Bayesian Knowledge Tracing (BKT) models were in active use in the Intelligent Tutoring Systems (ITS) field for over 20 years. They have been intensively studied, and a number of useful extensions to them were proposed and experimentally tested. Among the most widely researched extensions to BKT models are various types of individualization. Individualization, broadly defined, is a way to account for variability in students that are working with the ITS that uses BKT model to represent and track student learning. One of the approaches to individualizing BKT is to split its parameters into per-skill and per-student components. In this work, we are proposing an approach to individualizing BKT that is based on Hierarchical Bayesian Models (HBM) and, in addition to capturing student-level variability in the data, weighs the contribution of per-student and per-skill effects to the overall variance in the data. [For the full proceedings, see ED592609.]

Online planning of good teaching sequences has the potential to provide a truly personalized teaching experience with a huge impact on the motivation and learning of students. In this work we compare two main approaches to achieve such a goal, POMDPs that can find an optimal long-term path, and Multi-armed bandits that optimize policies locally and greedily but that are computationally more efficient while requiring a simpler learner model. Even with the availability of data from several tutoring systems, it is never possible to have a highly accurate student model or one that is tuned for each particular student. We study what is the impact of the quality of the student model on the final results obtained with the two algorithms. Our hypothesis is that the higher flexibility of multi-armed bandits in terms of the complexity and precision of the student model will compensate for the lack of longer term planning featured in POMDPs. We present several simulated results showing the limits and robustness of each approach and a comparison of heterogeneous populations of students. [For the full proceedings, see ED592609.]

The 9th International Conference on Educational Data Mining (EDM 2016) is held under the auspices of the International Educational Data Mining Society at the Sheraton Raleigh Hotel, in downtown Raleigh, North Carolina, in the USA. The conference, held June 29-July 2, 2016, follows the eight previous editions (Madrid 2015, London 2014, Memphis 2013, Chania 2012, Eindhoven 2011, Pittsburgh 2010, Cordoba 2009 and Montreal 2008). The EDM conference is the leading international forum for high-quality research that leverages educational data, learning analytics, and machine learning to answer research questions that shed light on the learning processes. This year's conference features three invited talks by: Rakesh Agrawal, President and Founder of Data Insights Laboratories; Marcia C. Linn, Professor of the University of California at Berkeley; and Judy Kay, Professor of the University of Sydney. Judy Kay's invited paper entitled "Enabling people to harness and control EDM for lifelong, life-wide learning" is also presented in the proceedings. Together with the "Journal of Educational Data Mining" ("JEDM"), the EDM 2016 conference supports a "JEDM" Track that provides researchers a venue to deliver more substantial mature work than is possible in a conference proceedings and to present their work to a live audience. The papers submitted to this track followed the "JEDM" peer review process; three papers have been accepted to the track and were presented at the conference. The abstracts of the invited talks, panels and accepted "JEDM" Track papers can be found in these proceedings. [For the 2015 proceedings, see ED560503.]

Bayesian Knowledge Tracing (BKT) has been in wide use for modeling student skill acquisition in Intelligent Tutoring Systems (ITS). BKT tracks and updates student's latent mastery of a skill as a probability distribution of a binary variable. BKT does so by accounting for observed student successes in applying the skill correctly, where success is also treated as a binary variable. While the BKT served the ITS community well, representing both the latent state and the observed performance as binary variables is, nevertheless, a simplification. In addition, BKT as a two-state and two-observation first-order HMM is prone to noise in the data. In this paper, we present work that uses feature compensation and model compensation paradigms in an attempt to conceptualize a more flexible and robust BKT model. Validation of this approach on the KDD Cup 2010 data shows a tangible boost in model accuracy well over the improvements reported in the literature. [This research paper was partially supported by the Andrew Mellon Pre-doctoral Fellowship and extends a project initiated during the Pittsburgh Science of Learning Center's Summer School at Carnegie Mellon University. For complete proceedings, see ED560503.]

Additive Factors Model (AFM) and Performance Factors Analysis (PFA) are two popular models of student learning that employ logistic regression to estimate parameters and predict performance. This is in contrast to Bayesian Knowledge Tracing (BKT) which uses a Hidden Markov Model formalism. While all three models tend to make similar predictions, they differ in their parameterization of student learning. One key difference is that BKT has parameters for the slipping rates of learned skills, whereas the logistic models do not. Thus, the logistic models assume that as students get more practice their probability of correctly answering monotonically converges to 100%, whereas BKT allows monotonic convergence to lower probabilities. In this paper, we present a novel modification of logistic regression that allows it to account for situations resulting in false negative student actions (e.g., slipping on known skills). We apply this new regression approach to create two new methods AFM+Slip and PFA+Slip and compare the performance of these new models to traditional AFM, PFA, and BKT. We find that across five datasets the new slipping models have the highest accuracy on 10-fold cross validation. We also find evidence that the slip parameters better enable the logistic models to fit steep learning rates, rather than better fitting the tail of learning curves as we expected. Lastly, we explore the use of high slip values as an indicator of skills that might benefit from skill label refinement. We find that after refining the skill model for one dataset using this approach the traditional model fit improved to be on par with the slip model. Parameter Estimation is appended. [For complete proceedings, see ED560503.]

In this paper, reinforcement learning (RL) algorithms Proximal Policy Optimization (PPO) and Deep Q Networks (DQN) are applied into the setting of autonomous driving research. A highway environment is formalized into a Markov Decision Process, and a simulated highway environment is constructed in compatible with Gymnasium. In this simulated environment, comparisons of PPO and DQN algorithm performance are investigated under different traffic load and vehicle starting position - main road and onramp - in terms of total return gained during an episode of highway driving after training all algorithms for the same number of timesteps. The total return takes driving speed and safety into account to connect with real world driving convention. According to the result, PPO algorithms achieve higher average return and lower standard deviation, indicating their ability to achieve more stable high-quality performance. All trained algorithms are able to drive quickly on the road and avoid other vehicles, demonstrating the potential of reinforcement learning in the application of complete driving automation. [ABSTRACT FROM AUTHOR]

Markov chain study is a widely used modeling tool that helps in understanding the change dynamics of land use land cover. This is an essential tool required for urban planning, its periodical monitoring as well as environmental planning, its protection, and its conservation. Through this study, an attempt has been made to understand the future dynamics of land use land cover of Guwahati city using multi-layer perceptron-Markov chain analysis (MLP-MCA). The past land cover maps of 1991 and 2001 were calibrated to predict for the year 2011. Subsequently, the predicted result of 2011 was validated with an observed land cover map of 2011 through ROC analysis. Finally, future land use land cover for the years 2031 and 2041 were predicted. The area statistics of 2031 and 2041 indicate that, there may be an escalation in built-up to the extent of 8.7 % by 2031 and 12.5 % by 2041 with significant deterioration in natural and semi-natural resources in the region. [ABSTRACT FROM AUTHOR]

The concept of forecasting with fuzzy time series have been increasingly developed to solve various problems. The length of the interval and defuzzification are two important factors that affect the accuracy of forecasting results. In this paper, the author combines the fuzzy time series-Markov chain with particle swarm optimization algorithm. Markov chain rule was implemented in the defuzzification for handling the repeated fuzzy sets and determining the proper weights. Meanwhile, the particle swarm optimization was applied for determining the appropriate length of intervals of the fuzzy time series by considering the universe of discourse as a search space and intervals as particles. The author uses the Average Forecasting Error Rate (AFER) value to see the level of accuracy of forecasting. The proposed method was applied to predict the opening stock price of PT. Astra International Tbk. and showed a very good accuracy with an AFER value of 0.9555%. [ABSTRACT FROM AUTHOR]

An equation is derived for determining the characteristic exponents of a system of linear differential equations with coefficients depending on a random Markov periodic process at parametric resonance and using it, sufficient conditions for the existence of an unstable solution to the system are derived. [ABSTRACT FROM AUTHOR]

The financial crisis is caused by the overbalance of the exchange rate of domestic currency against foreign currency. The 1997/1998 ASIAN financial crisis and the 2008 global financial crisis have provided a considerable impact on the Indonesian economy. This study aims to construct a time series model that can detect financial crisis in Indonesia using combination of Markov Switching (MS), Dynamic Conditional Correlation (DCC), and Generalized Autoregressive Conditional Heteroscedasticity (GARCH). In this study we consider the difference in real BI rate with real Fed rate and deposit rate as indicators to observe the changes of financial crisis. The indicator of the difference in the real BI rate with the real Fed rate and the real deposit interest rate is detected to have a causality relationship and is cointegrated so that it can be explained using the multivariate vector error correction (VEC) model. The VEC model contains heteroscedasticity effects that can be overcome by using a volatility model. The method used for the detection of financial crisis was a combination of volatility models and Markov switching. The results showed that the Markov switching model dynamic conditional correlation generalized autoregressive conditional heteroscedasticity or MS-DCC-GARCH (2,1,1) can detect the crisis in 1997/1998. Based on the predicted value of smoothed probability in 2020, there was no sign of financial crisis in Indonesia. [ABSTRACT FROM AUTHOR]

Public interest in investment instruments such as gold and stocks has increased in the last four years. This study used the Markov Chain to compare the performance of two investment instruments: gold and stocks. Although many s udies have used the Markov Chain to predict investment performance, no research compares different types of instruments simultaneously. On the other hand, there has been no research linking the results to the Covid-19 pandemic. This study used the Markov Chain because it is accurate and suitable for forecasting stock prices and gold. The results show that five stock sectors will experience an increase with the probability of more than 60%: transportation, consumer cyclical, energy, finance, and industrial. On the other hand, gold will also experience an increase in the long term with the probability of more than 50%. Markov chain can be used as the initial phase by selecting sectors likely to increase before urther technical analysis. First passage time indicates that stocks are more volatile, with an average transition from "up" to "down" for only one month. [ABSTRACT FROM AUTHOR]

Crop yield forecasting is the social problem in a current situation of the weather condition. Farmers' livelihoods are critically dependent on a changing environment that they are failing to fathom without access to sophisticated farming techniques and science-based climate and weather data. To deal with this case and to create the attention and awareness to the farmers, this work attempts to seek out the equilibrium condition for the weather to predict and to detect the crop yield forecasting using hidden markov model. Neutrosophic set is that the universal method to deal with indeterminacy. Markov process could be a stochastic model for estimating the equilibrium of any system, whereas in a hidden markov model the current state emits an observation. This work proposes and applies a proposed concept in agricultural field forecasting utilising a hidden markov model with a single valued neutrosophic number. [ABSTRACT FROM AUTHOR]

Rectifiers are unavoidable parts of most of the electronic circuits which necessitate a proper analysis and understanding of the ripple component present at the output of the rectifier. The present study employs a unique methodology for studying the ripple component through the construction of a complex network utilizing the Markov model. The analysis of the Markov chain reveals that a drop in the ripple component is associated with the rise in number of disconnected nodes and last node's self-loop probability. In the absence of filtering, it is observed that the output of the rectifier exhibits interconnectivity among all the nodes through edges. There is a positive correlation between filtering efficiency and indegree. [ABSTRACT FROM AUTHOR]

Markov models are frequently employed in a variety of real-world systems, including industrial and queuing systems. Higher order Markov model (HOM) of inventory system including remanufacture, substitution, and lost sales is developed in this study. Additionally taken into consideration are deterioration during storage, disposal, and probabilistic demand. A separate warranty period for newly created and previously manufactured goods is returned to the manufacturer. We believe that a higher order Markov model has only occasionally been applied to a production inventory system. Finding the system's steady state presents a hurdle. Another challenge is that one of the state variables depends not on its previous states but on the previous states of other variables. determines the system's requirements at its steady state 3910 regulations. Our results show increasing in the re-manufactured and deteriorated items over time. Also, inventory levels in the three warehouses. Meanwhile, an opposite conclusion in the case of manufactured, substitution, and lost sales. Substitution processes lead to increase the expected average profit. [ABSTRACT FROM AUTHOR]

User behavior is the main factor for evaluating any system, and based on this, the structure is optimized and the system is reorganized to provide high quality services to the users. This work is dedicated to analyzing the dynamics of user behavior in e-government using common techniques, Customer behavior model graphs (CBMG) and Markov chains. The analysis of user behavior showed that, a lot of time is spent on searching for necessary services, browsing information, and filling out corresponding forms and in order for successful use of e-government systems in our country, it is necessary to simplify the process of filling out various forms and increase the level of automation when receiving government services. [ABSTRACT FROM AUTHOR]

The article presents the experience of building a system for early prediction of learning success based on the Markov model. The simulation is based on data from the LMS and the solution of the Kolmogorov system of equations. The advantage of this model is that it makes it possible to solve the problems of managing the educational process in real time. The difference between the model proposed in the work and most works, in which learning success is considered from the point of view of predicting student dropouts, is the early prediction of learning success in a particular discipline. [ABSTRACT FROM AUTHOR]

This paper examines the effect of real-time, personalized alignment of a robot's reward function to the human's values on trust and team performance. We present and compare three distinct robot interaction strategies: a non-learner strategy where the robot presumes the human's reward function mirrors its own; a non-adaptive-learner strategy in which the robot learns the human's reward function for trust estimation and human behavior modeling, but still optimizes its own reward function; and an adaptive-learner strategy in which the robot learns the human's reward function and adopts it as its own. Two human-subject experiments with a total number of N=54 participants were conducted. In both experiments, the human-robot team searches for potential threats in a town. The team sequentially goes through search sites to look for threats. We model the interaction between the human and the robot as a trust-aware Markov Decision Process (trust-aware MDP) and use Bayesian Inverse Reinforcement Learning (IRL) to estimate the reward weights of the human as they interact with the robot. In Experiment 1, we start our learning algorithm with an informed prior of the human's values/goals. In Experiment 2, we start the learning algorithm with an uninformed prior. Results indicate that when starting with a good informed prior, personalized value alignment does not seem to benefit trust or team performance. On the other hand, when an informed prior is unavailable, alignment to the human's values leads to high trust and higher perceived performance while maintaining the same objective team performance. [ABSTRACT FROM AUTHOR]

A Markov Chain Monte Carlo method (MCMC) is adopted to estimate a continuous hidden Markov autoregressive model; best fitted to formulate the hidden behavior of a disease progression. The motivation behind this work is to unravel the latent disease stages during its development in some pathologies such as cancer, where during treatment the disease stage is not known directly unless we observe other parameters like bio-markers. By the way, we suppose we have individual patient observations (bio-markers'measurements) with different number of observation points and with non equidistant time intervals, hence we work in a continuous time framework. Usually, the likelihood is unavailable in analytic form for most of the Markov switching models (MSM); particularly that the hidden states are unknown parameters and by the way the likelihood is augmented with the hidden states as new parameters. To better deal with the augmented parameters and to use a likelihood up to constant, we call for the Bayesian MCMC tools by using data augmentation in the likelihood; which would alternate between estimating the parameters and evaluating the hidden states. The posterior densities for the parameters will be developed after prior specifications, and the hidden states will be computed by Gibbs Sampler block update using a forward filtering backward smoothing (FFBS) scheme. Finally, the estimation of transition rates between states is well explained as well as the MCMC algorithm results are provided. [ABSTRACT FROM AUTHOR]

As COVID-19 continually to invade all the continents globally, a projection of the spread of COVID-19 is crucial in combating the outbreak. Thus, the main objective of this research is to forecast the spread of COVID-19 pandemic in Johor at sub-district level. The daily cases of COVID-19 for 190 days in 92 sub-districts in Johor were collected. The COVID-19 situation in each of the sub-districts was classified into three distinct state spaces, which are the green zone (zero reported cases per day), alert zone (one to seven reported cases per day), and the danger zone (more than seven reported cases per day). The Markov chain method is adopted to get the stationary distribution for all the 92 sub-districts in Johor. Most of the sub-districts are predicted to be in an alert zone in the long term, that is, have a high chance of getting 1 to 7 reported cases per day. There are 25 sub-districts in Johor that are expected to have zero reported cases per day in the long term. The rest of the sub-districts in Johor are predicted to be in a danger zone in the long run, that is, a high chance of observing more than 7 reported cases per day. [ABSTRACT FROM AUTHOR]

Almost all motor insurance uses No-Claim Discount (NCD) system in which the premium depends on the level at which the policyholder belongs. At the end of a claim-free year the policyholder moves to a level with a lower premium or stays on the level he/she is currently at. On the contrary, if the policyholder made at-fault accident(s) in the previous year, he/she moves to a level with a higher premium. In this case, the policyholder is penalised by surcharges. This system can be modelled using Markov chains. Transitions from one premium level to another are determined once the number of claims is known. However, we should be aware that policyholders might decide not to report small claims to the insurance company if they believe that this is profitable for them. Such behaviour is called bonus hunger. This research examines the effect of bonus hunger on the five NCD systems namely Malaysia, Thailand, the United Kingdom, the Co-Operative Insurance and Brazil using Markov Chains processes and several actuarial tools. It is found that Malaysia has the toughest system, but it has the best spread of policyholders between the premium classes, and it is the most responsive to a change in the number of claims. Nonetheless, the presence of bonus hunger makes the main objective of the NCD system less achievable. The system is unable to punish bad drivers as they do not report the claims and thus, resulting in less sensitivity to claim frequency and even higher concentration of policyholders in lower premium classes. [ABSTRACT FROM AUTHOR]

This work will give a novel approach in predicting premium of Indonesian's health insurance for the next 30 years using application of dynamical systems. First, we fit Indonesian population dataset from 1960–2018 and the number of policy holders to estimate the coefficients of our dynamical systems using Markov Chain Monte Carlo Process. We used models fit to predict the number of participant for the next 30 years. We then implemented the frequency-severity model to evaluate the premium of each segment in the insurance policy. The projected premium for the next 30 years will be calculated using the same frequency-severity model but with new parameters obtained from the numerical solution of the dynamical system. We assumed a constant proportion between population data to the number of participant in the sample data and a gradually decreasing proportion between the number of zero claim and the number of sample data. Premium in all segments overestimated costs regulated by the government with one exception in one segment. [ABSTRACT FROM AUTHOR]

The study analyzes and assess the traditional generalized autoregressive conditional heteroscedasticity (GARCH) model with the Markov regime-switching GARCH model (MRS GARCH). Evaluation based on the accuracy of forecasting volatility and risk of Malaysia natural rubber grade Standard Malaysia Rubber 20 (SMR20). We fitted these models under six distributions which are normal, Student-t, generalized error distribution (GED), and their skewed version. Based on log-likelihood (LL), Akaike information criterion (AIC), and Bayesian information criterion (BIC), we found that the MRS GARCH models outperform the traditional GARCH models. We also found that the SMR20 returns are fat tails and skewedly distributed. Further, we forecasted one-day Value-at-Risk (VaR) for each model and compared their adequacy using the conditional coverage (CC) test and Dynamic Quantile (DQ) test. We discovered that the best accurate VaR prediction for SMR20 risk management comes from the MRS GARCH model of skewed Student-t distribution. [ABSTRACT FROM AUTHOR]