Adaptive Learning System with Educational Data Mining and Learning Analytics

Abstract

As per the 21st century educational trends, there is increase in the size of the cohort of students in the present classrooms. Further, present COVID pandemic situation, switching the modes of learning from offline to online, and inclusion of technology in the existing learning environments, the fields of educational data mining, learning analytics and adaptive learning systems have gained much attention worldwide. These help in optimizing the educational environments in various ways like providing personalized learning support, improving student learning outcomes, instructor performance and curriculum, identifying career learning pathways etc. This research study aims to develop an adaptive learning system with educational data mining and learning analytics, for the blended learning environments. To develop the proposed system, the distinguishment between the knowledge level of the students is supported. Since the study is performed for blended learning environments, the viewpoints, characteristics, behavioural analysis and identification of probable parameters for different category students (high versus low performers) is necessary. This research study helps to build an understanding of low performer and at-risk (failure or dropout) students’ behaviour by interviewing low performer, at-risk students of fifth semester undergraduate engineering students. It focuses on the underlying reasons behind their low performance, at-risk behaviour for blended learning environments. This study explores their learning behaviour and perceptions, reasons for being low performers and objectives behind studying the courses. The research findings reveal that majority students belonging to the at-risk, low-performer category are found to be well-aware about their present academic state. They intentionally devote a minimal time to the courses, as they are focusing on their other prioritized aspects like building of new skills-set for making themselves market ready. Through these research findings, this study also proposes flexible design and reflections for blended learning environments. For further exploring the reliability, behavioural analysis of acquired feedback, this research work also addresses the question of whether gender and socio-economic differences affect students’ opinion of their teachers in higher education, across a number of disciplines. The research analyzed the differentials in students’ ratings of their teachers in five disciplines in the field of education. Data was drawn from student responses to the surveys conducted in a large Indian university at the end of each course unit. This study analyzes 112919 and 16354 complete sets of student ratings, to study the gender and socio-economic diversity based effects respectively. Statistical multivariate and univariate general linear models were used to derive the relevant results and graphs. The study reveals the existence of socio-economic status bias, gender-typical behaviour, gender-atypical behaviour, and same-gender and cross-gender biases; these resulted in differential ratings in the disciplines examined. For efficient designing of the proposed system for blended learning environments, this research study also focusses on exploring the workable parameters for available online environments. This study explores two types of logs (activity and performance) for identification of reliable parameters. For first study, it contributes towards efficient at-risk student identification by proposing a generalized sequential deep learning model for learning behaviour identification, built upon temporal 3-dimensional clickstream data. The model is trained and validated on two big public datasets: KDD Cup 2015 and OULAD. The results on the unseen test data achieve a classification accuracy ranging from 87.6%-92% and AUC from 0.881-0.961 and outperform other existing baseline models. For performance parameters based study, it proposes a modelling methodology for deploying interpretable Hidden Markov Model for mining of the sequential learning behaviour from light-weight assessments. The public OULAD dataset having diversified courses and 32,593 student records is used for validation. The results on the unseen test data, achieve a classification accuracy ranging from 87.67%-94.83% and AUC from 0.927-0.989, and outperforms other baseline models. For implementation of early warning systems the study also predicts the optimal time period, during the first and second quarter of the courses. With the outcomes, this study tries to establish an efficient generalized modelling framework that may lead the higher educational institutes towards sustainable development. After the identification of performance parameters as reliable inputs, the research work focusses on checking the effectivity of these for blended learning environments. This research work utilizes lightweight formative assessments for blended learning environments. It discusses their implementation and effectiveness in early identification of at-risk students. This study validates the usage of lightweight assessments for three core pedagogically different courses of large computer science engineering classrooms. It uses voting ensemble classifier for effective predictions. With the usage of lightweight assessments in early identification of at-risk students, accuracy range of 87%–94.7% have been achieved along-with high ROC-AUC values. The study also proposes the generalized pedagogical architecture for fitting in these lightweight assessments within the course curriculum of pedagogically different courses. With the constructive outcomes, the light-weight assessments seem to be promising for efficient handling of scaling technical classrooms. At the end, as an outcome of this research work, it presents an adaptive learning system with educational data mining and learning analytics. The proposed system is built upon topic based learning management system with various basic functionalities. The web interface provides support for student and teachers in effective management of the course. For research purpose, the topics are mapped with the course knowledge components. These are used to trace the student knowledge level. The adaptive interface with dashboard helps to judge the knowledge level of the student and provides adaptive educational content recommendations. For support of educational data mining and learning analytics, it is integrated with a machine learning portal. This portal helps the instructors in easy analysis. The instructors can train, test course specific models. They can also make early predictions regarding the student at-risk behaviour and can provide timely intervention to the respective students. This proposed system has the capability to embed effectiveness in the blended learning environments and can be used for the benefits of the society in the educational field.