PhD / HDR defences

PhD defence, Dynil DUCH

Date : 21/05/2025
Time : 9h00
Location : Institut d’Informatique Claude Chappe, Amphitheatre Hall

Phd defence video :

Photo significative du sujet de thèse

 

Title – Predicting Student Performance through Cross-Institutional Learning Analytics: CISE Model and Reflective Learning Tools

 

Jury composition :

  • M. Yvan PETER – University Professor, Université de Lille (President)
  • Mme Armelle BRUN – University Professor, Université de Nancy (Reporter)
  • Mme Agathe MERCERON – University Professor, Université de Berlin (Reporter)
  • M. François BOUCHET – Senior Lecturer, Sorbonne Université (Reviewer)
  • Sébastien GEORGE, Professor, Le Mans Université (Thesis director)
  • Madeth MAY, Senior Lecturer, Le Mans Université (Thesis co-supervisor)

 

 

Abstract :

The rapid evolution of digital learning environments has underscored the need for innovative tools to enhance student performance, engagement, and equity. This dissertation introduces two frameworks: the Cross-Institutional Stacking Ensemble (CISE) predictive model and ReflectMate, a reflective learning analytics tool. Together, they address challenges in ICT education, such as dataset heterogeneity, class imbalance, and disparities caused by the digital divide, transforming educational practices through predictive analytics and reflective learning.
The CISE model is an ensemble learning framework that forecasts student academic performance across diverse institutional contexts. It integrates five machine learning classifiers—Decision Tree, Random Forest, Naive Bayes, Neural Network, and Support Vector Machine—synthesized via a logistic regression-based meta-model. Achieving an F1 score of 78.25% on the ENSIM validation dataset, CISE demonstrates superior accuracy and generalizability in identifying at-risk students.
Complementing CISE, ReflectMate empowers students with real-time insights into their learning behaviors (e.g., quiz attempts, assignment submissions, LMS interactions). Through progress tracking, tailored feedback, and interactive dashboards, ReflectMate fosters self-reflection, self-regulation, and accountability. Surveys conducted among CADT students confirm its effectiveness in promoting metacognition and academic engagement.
This research also addresses digital equity, revealing gender and location-based disparities in ICT education. By combining predictive modeling with reflective learning tools, this work contributes scalable, student-centered solutions to bridge systemic gaps and promote inclusive academic success.
Keywords : Student performance prediction, learning analytics, predictive modeling, reflective learning, cross-institutional validation, student engagement.

Photo après soutenance