Program

Explore the accepted sessions for The Learning Ideas Conference 2026 below!

Our program will also include a featured panel discussion and keynote talks from:

  • Dr. Maciej Pankiewicz, Senior Research Investigator and Associate Director at the Penn Center for Learning Analytics, University of Pennsylvania

  • Megan Torrance, CEO of TorranceLearning

  • Dr. Candace Thille, Associate Professor and Faculty Director for Adult and Workforce Learning at the Stanford Accelerator for Learning, Stanford University

The complete conference program, including session times, will be published in April.

Page 1 | Page 2 | Page 3 | Page 4

Identifying At-Risk Students: An Explainable, Actionable, and Hybrid Approach Using Machine Learning and Large Language Models

Sherif Abdelhamid, Ph.D., and Mona Aly, Virginia Military Institute, Lexington, Virginia, USA

Student dropout remains a critical challenge in higher education, leading to substantial academic, financial, and societal consequences. While prior and current research has explored artificial intelligence techniques for predicting student dropout, most existing studies focus primarily on standalone predictive models, offering limited support for real-time decision-making, explainability, and actionable interventions. Moreover, the use of large language models (LLMs) to analyze unstructured student data and complement quantitative predictions remains largely unexplored in this domain. These gaps motivate the need for integrated, intelligent systems that not only predict dropout risk but also help explain underlying causes and support proactive interventions.

This research addresses the problem of identifying and predicting student dropouts, framing it as a classification task to spot at-risk students early in their academic journey…

Keywords: Student Dropout Prediction, Machine Learning in Education, Ensemble Learning, Large Language Models (LLMs), Learning Analytics and Decision Support Systems

Identifying At-Risk Students: An Explainable, Actionable, and Hybrid Approach Using Machine Learning and Large Language Models

Sherif Abdelhamid, Ph.D. and Mona Aly

Student dropout remains a critical challenge in higher education, leading to substantial academic, financial, and societal consequences. While prior and current research has explored artificial intelligence techniques for predicting student dropout, most existing studies focus primarily on standalone predictive models, offering limited support for real-time decision-making, explainability, and actionable interventions. Moreover, the use of large language models (LLMs) to analyze unstructured student data and complement quantitative predictions remains largely unexplored in this domain. These gaps motivate the need for integrated, intelligent systems that not only predict dropout risk but also help explain underlying causes and support proactive interventions.

This research addresses the problem of identifying and predicting student dropouts, framing it as a classification task to spot at-risk students early in their academic journey. We conduct a comprehensive machine learning study using a publicly available student performance dataset from a public repository, which includes demographic, academic, financial, and socioeconomic features. After data preprocessing and feature scaling, and an 80/20 train–test split, multiple machine learning models are evaluated, including Logistic Regression, Decision Trees, Support Vector Machines, K-Nearest Neighbors, Random Forests, Feedforward Neural Network, Gradient Boosting, XGBoost, and ensemble methods (majority voting and stacking). Experimental results show that a majority-voting ensemble classifier achieves the best performance, with an accuracy of 94.9%, along with strong precision (95.3%), recall (91%), and F1-score (93%).

Beyond model development, this study presents EduPulse, an AI-powered web-based platform that integrates our best-performing machine learning model with an LLM (Claude Sonnet 4.5). EduPulse provides institution-, school-, and student-level dashboards, advanced filtering, and a what-if simulation module to assess the impact of academic and financial interventions. The LLM component analyzes unstructured student notes and records to extract sentiment and dropout-related indicators, offering qualitative insights that enhance model interpretability and decision support.

To the best of our knowledge, this work is among the first few studies to integrate ensemble machine learning and large language models within a unified platform for student dropout prediction and intervention. The results demonstrate that combining predictive analytics with LLM-based qualitative analysis enables accurate, interpretable, and actionable insights, supporting data-driven strategies to improve student retention and success in higher education.

 
IGIP SESSION

A Methodology for Teaching Economics in the Digital Era

Galiya Berdykulova, Ph.D., International IT University, Almaty, Kazakhstan

The underestimation of scientific advances in development theories related to post-industrial society and breakthrough innovations as requirements of a new civilization and a new paradigm generated by the digital age is a problem of standard curricula and disciplinary programs. This session is related to the need to find out how changes in economic science should be reflected in the content of economic disciplines such as economic theory and economics and industrial engineering. One of the ways to achieve balance and harmonization of science and educational practice is to update the teaching methods of economic disciplines.

A review of relevant literature, original examples of post-industrial society and breakthrough innovations in the context of digitalization in Kazakhstan, the principle of concreteness and the principle of scientific knowledge were used to find ways to eliminate the undervaluation of new knowledge in the teaching of economic theory and the disciplines of economic and industrial engineering…

Keywords: teaching methodology, digitalization, principle of specificity, principle of scientific knowledge

A Methodology for Teaching Economics in the Digital Era

Galiya Berdykulova, Ph.D.

The underestimation of scientific advances in development theories related to post-industrial society and breakthrough innovations as requirements of a new civilization and a new paradigm generated by the digital age is a problem of standard curricula and disciplinary programs. This session is related to the need to find out how changes in economic science should be reflected in the content of economic disciplines such as economic theory and economics and industrial engineering. One of the ways to achieve balance and harmonization of science and educational practice is to update the teaching methods of economic disciplines.

A review of relevant literature, original examples of post-industrial society and breakthrough innovations in the context of digitalization in Kazakhstan, the principle of concreteness and the principle of scientific knowledge were used to find ways to eliminate the undervaluation of new knowledge in the teaching of economic theory and the disciplines of economic and industrial engineering. Learning theory, behavioral science, and communication management theory with a rhetorical research methodological model were implemented to overcome the negative impact of disruptive innovations in the digitalization of education. To find new insights into the teaching of economic theory and the disciplines of economics and industrial organization, a review of relevant literature, original examples of post-industrial society and breakthrough innovations in the digitalization of Kazakhstan, principles of concreteness, and principles of scientific knowledge were used.