Source-linked AI summary
Educational data mining and learning analytics: An updated survey
C. Romero, S. Ventura
TL;DR
Educational Data Mining and Learning Analytics face rapidly expanding educational data and a field that has changed substantially since the previous survey. This paper updates the survey through a broad review of the field’s knowledge, environments, tools, datasets, methods, objectives, and future trends, showing that the area now encompasses more terminology, applications, resources, and emerging environments.
Problem
Rapidly expanding educational data and substantial growth in EDM and LA literature have created a need for an updated, comprehensible overview of the field.
Method
The paper conducts a broad survey of EDM and LA, including their knowledge discovery cycle, educational environments, tools, datasets, methods, objectives, and future trends.
Results
The updated survey finds expanded terminology, publications, educational environments, tools, free datasets, application topics, and future trends in EDM and LA.
Takeaways & Limitations
EDM and LA now span a broader and more varied research landscape involving new environments, resources, application problems, and emerging trends.
Takeaways & Limitations
The survey’s taxonomy does not cover all possible EDM and LA tasks or objectives across the many stakeholder groups.
Abstract
from arXiv · showhide
This survey is an updated and improved version of the previous one published in 2013 in this journal with the title data mining in education. It reviews in a comprehensible and very general way how Educational Data Mining and Learning Analytics have been applied over educational data. In the last decade, this research area has evolved enormously and a wide range of related terms are now used in the bibliography such as Academic Analytics, Institutional Analytics, Teaching Analytics, Data-Driven Education, Data-Driven Decision-Making in Education, Big Data in Education, and Educational Data Science. This paper provides the current state of the art by reviewing the main publications, the key milestones, the knowledge discovery cycle, the main educational environments, the specific tools, the free available datasets, the most used methods, the main objectives, and the future trends in this research area.
First author: Full name and affiliation; plus email address if corresponding author
The supplied passage lists research-area terms but does not provide the first author’s name, affiliation, or email address.
- The passage names Educational Data Mining, Learning Analytics, and related education data-science terms.It lists Data Mining on Education, Data-Driven Decision Making in Education, Big Data in Education, and Educational Data Science.
INTRODUCTION
The growth of educational data has created a central challenge: transforming large repositories into insights that benefit students, teachers, and administrators. This survey updates a 2013 overview by synthesizing developments in EDM and LA, related terminology, educational environments, tools, datasets, applications, and future trends.
- Motivation: Educational institutions face exponentially growing educational data and the challenge of transforming it into useful insights.These repositories arise from e-learning resources, educational software, Internet use, and state student-information databases.
- EDM and LA: Educational Data Mining develops methods for exploring educational-environment data and applying data-mining techniques to important educational questions.Learning Analytics measures, collects, analyzes, and reports learner and contextual data to understand and optimize learning environments.
- EDM and LA: EDM and LA share data-intensive educational research and practice goals, while LA emphasizes data-driven decisions and social-pedagogical integration and EDM emphasizes technological challenges.Both are interdisciplinary combinations of computer science, education, and statistics.
- Related terminology: The bibliography also uses terms including Academic Analytics, Institutional Analytics, Teaching Analytics, Data-Driven Education, Data-Driven Decision Making, Big Data in Education, and Educational Data Science.These terms address institutional insight, teaching activities, educational decisions, big-data techniques, and educational problem solving.
- Survey contribution: This survey updates the 2013 Data Mining in Education review because six years of research produced many new papers and substantial developments across the field.Changes include increased LA interest, new educational environments, more specific tools and free datasets, broader application topics, and new future trends.
BACKGROUND
EDM and LA developed through distinct conference communities and an expanding body of books and journals. Interest grew sharply over the last two decades, with LA surpassing EDM in references after 2011.
- Conferences: EDM and LA emerged from independent communities, marked by inaugural conferences in 2008 and 2011, respectively.The first EDM conference was held in Montreal and the first LAK conference in Banff.
- Books: The field’s book literature expanded from the 2006 volume Data Mining in E-Learning to later works emphasizing learning analytics, data science, and big data.Titles shifted from Data Mining in Education and Educational Data Mining during 2006–2014 toward newer terminology during 2015–2017.
- Journals: Two specialized journals anchor EDM/LA publishing: the Journal of Educational Data Mining launched in 2009 and the Journal of Learning Analytics in 2014.The survey also notes newly established related journals, including the International Journal of Learning Analytics and Artificial Intelligence for Education.
- Research growth: Search results for “Educational Data Mining” and “Learning Analytics” grew exponentially from 2000 to 2018, with LA surpassing EDM after 2011.EDM had more references than LA until 2011, after which LA became more prevalent.
- Influential research: Four in ten of the most cited EDM and LA papers are reviews or surveys, beginning with Romero and Ventura’s 2007 review.Later influential reviews by the same authors appeared in 2010 and 2013.
EDM/LA KNOWLEDGE DISCOVERY CYCLE
EDM/LA applies a knowledge discovery cycle based on the general KDD process, with each step adapted to educational data and environments. The cycle gathers and preprocesses heterogeneous data, applies mining methods, and uses comprehensible results to support educational interventions and decisions.
- Cycle foundation: EDM/LA follows the general Knowledge Discovery and Data Mining process, while each cycle step has specific educational characteristics.The process is presented as a cycle application of KDD with important differences in each step.
- Data collection: Educational environments and information systems determine which data can be collected to address different educational problems.Examples include traditional, computer-based, and blended education, as well as LMS, ITS, and MOOC systems.
- Data collection: Educational data combine interaction, administrative, demographic, and affective information gathered from multiple sources.Examples include navigation behavior, quiz inputs, forum messages, school and teacher information, gender, age, motivation, and emotional states.
- Data preprocessing: Data preprocessing is difficult because raw educational data usually require conversion into an appropriate form for each problem.Preprocessing can take more than half of the total time spent solving a data mining problem, while ethical issues include privacy and informed consent.
- Mining methods: Traditional techniques such as visualization, classification, clustering, and association analysis have been applied successfully, but hierarchical and longitudinal educational data require specific treatment.EDM and LA use a wide range of methods and techniques for different educational problems.
- Action and interpretation: The cycle ultimately requires instructors and academic authorities to use discovered knowledge for interventions and decisions that improve student learning performance.Results must be comprehensible for decision-making; visualization and recommender systems can make findings easier to interpret and provide explanations or recommendations.
EDUCATIONAL ENVIRONMENTS AND DATA
Educational data arise from diverse environments, including traditional, computer-based, and blended learning systems. These environments collect different types of information, from attendance and marks to data generated through face-to-face and computer-mediated instruction.
- Educational environments include traditional education, computer-based education, and blended learning, each providing different data sources.
- Traditional education: Traditional education is primarily face-to-face and collects student attendance, marks, curriculum goals, class, and schedule information.
- Computer-based education: Computer-based educational systems include Adaptive and Intelligent Hypermedia Systems, Intelligent Tutoring Systems, Learning Management Systems, and Massive Open Online Courses.
- Blended learning: Blended Learning environments combine face-to-face and computer-mediated instruction, increasing access, convenience, flexibility, and freedom by moving significant sessions online.
TOOLS AND DATASETS
EDM/LA research can draw on many general-purpose tools and some free public datasets, but educators face methodological barriers and researchers face costly data preparation. Public datasets remain limited and unevenly distributed across educational environments, motivating dedicated repositories and interoperability standards.
- Tools: General-purpose frameworks such as RapidMiner, Weka, SPSS, KNIME, Orange, and Spark Lib support EDM/LA research but require users to select algorithms and parameters.These requirements make the tools difficult for educators to use effectively.
- Datasets: Researchers often use their own data, although gathering and preprocessing educational data is hard and time consuming.Free public datasets provide an alternative source for EDM/LA studies.
- Datasets: DataShop is highlighted as one of the first and biggest datasets that also provides a tool for researching intelligent tutoring systems.It serves as a central repository for securing and storing research ITS data, with analysis and reporting tools.
- Datasets: Public EDM/LA datasets are few and do not cover all educational environments, with most originating from e-learning systems.The paper proposes an EDM/LA dataset repository analogous to the UCI Machine Learning Repository.
- Datasets: Interoperability could be improved through standards such as Experience API (xAPI) or IMS Caliper instead of incompatible “walled gardens.”The passage frames these standards as an alternative to systems that cannot work together.
METHODS AND APPLICATIONS
EDM/LA uses diverse general data-mining methods alongside education-specific techniques to address a broad range of educational problems. Applications span causal analysis, student modeling, prediction, recommendations, curriculum improvement, emotional and self-regulated learning, and personalized feedback.
- Methods: EDM/LA combines universal data-mining methods, including visualization, prediction, clustering, outlier detection, relationship mining, causal mining, social network analysis, process mining, and text mining, with education-specific techniques.Education-specific methods include distillation of data for human judgment, discovery with models, knowledge tracing, and non-negative matrix factorization.
- Methods: Causal mining identifies causal relationships or effects, clustering finds similar observations, prediction infers target variables, process mining learns from event logs, and recommendation predicts user preferences.These methods support applications such as identifying causes of learning outcomes, grouping students or materials, predicting performance and behaviors, analyzing educational processes, and recommending activities or courses.
- Methods: Knowledge tracing estimates student mastery from cognitive models and answer logs, while nonnegative matrix factorization decomposes test outcomes into item and student-mastery matrices for skill assessment.Knowledge tracing monitors student knowledge over time, and nonnegative matrix factorization supports assessment of student skills.
- Applications: EDM/LA objectives extend beyond learners and instructors to many stakeholders and include curriculum analytics, dashboards, early prediction of at-risk students, intervention evaluation, and interpretable learner models.Additional applications address serious-game interactions, foreign-language learning, self-regulated learning, causality discovery, and emotional learning analytics.
- Applications: Current applications also use machine learning and sensor technologies to orchestrate learning analytics, provide personalized feedback, and understand navigation paths.The research community additionally applies data-mining and visualization techniques to player interactions in serious games.
CONCLUSIONS AND FUTURE TRENDS
EDM and LA have expanded rapidly into established interdisciplinary research communities and toward broader market use. Future priorities include accessible general-purpose tools, responsible implementation, defined research challenges, and lifelong real-time educational analytics.
- EDM and LA have grown rapidly, with dedicated conferences and journals plus increasing books, papers, and surveys.
- Future trends: Specific-purpose tools are now widely available, but general-purpose EDM/LA tools remain a challenge for broader adoption.
- Future trends: LA implementations should ensure transparent goals, appropriate feedback environments, secure and private data practices, scalability, and beneficial outcomes for stakeholders.
- Future trends: BLAP identifies six challenges spanning transferability, effectiveness, interpretability, applicability, and two forms of generalizability.
- Future trends: AUC ROC >= 0.65 is required for a general-purpose boredom detector, while cross-population detectors should degrade by under 0.1 and remain better than chance.
- Future trends: Future EDM/LA may integrate students’ data across their lives and use big data, IoT, and real-time portable models to support educational systems and agents.