@ARTICLE{10.21494/ISTE.OP.2026.1447, 

TITLE={Hybrid Approach Combining Markov, HMM, and RNN to Detect Learning Blockages in Programming Résumé}, 
  
AUTHOR={Grota Abdelkader
, Mohammed Erritali
, Patrick Etcheverry
, Thierry Nodenot, }, 
	
JOURNAL={Open Journal in Information Systems Engineering}, 
	
VOLUME={6}, 

NUMBER={Special Issue INFORSID 2025}, 
	
YEAR={2026}, 

URL={https://www.openscience.fr/Hybrid-Approach-Combining-Markov-HMM-and-RNN-to-Detect-Learning-Blockages-in}, 
	
DOI={10.21494/ISTE.OP.2026.1447}, 
	
ISSN={2634-1468}, 
	
ABSTRACT={Understanding the learning process in programming poses a complex challenge due to the sequential and multidimensional nature of students’ interactions with digital environments. This study analyzes the activity logs of 70 beginner computer science students engaged in problem-solving programming tasks to identify encountered difficulties and propose tailored pedagogical interventions.We present a hybrid approach combining Markov Chains to model transitions between types of actions, Hidden Markov Models (HMM) to infer latent learning states (progress, hesitation, blockage), and Recurrent Neural Networks (RNN) enhanced with an attention mechanism to detect critical moments. This combination allows for the simultaneous exploitation of behavioral, cognitive, and sequential dimensions of learning.The methodology involves extracting temporal and structural features from the activity logs, modeling cycles of exploration, hesitation, and blockage, and integrating them into a unified framework to predict learning states. The data comes from several standardized practical work sessions, totaling over 80 MB of timestamped traces, collected with consent and stored in a NoSQL database. Each session is divided into sequences corresponding to coherent problem-solving phases, enabling a fine-grained analysis of learning trajectories.Experimental results show that the proposed hybrid model outperforms traditional approaches, achieving an accuracy of 93.5% and significantly reducing false positives in blockage detection.
The multidimensional analysis offers a richer understanding of learning trajectories, including their invisible dimensions such as unproductive engagement or phases of uncertainty.This research paves the way for intelligent educational platforms capable of providing personalized real-time feedback, sensitive to micro-indicators of activity, cognitive intensity, and individual context, thereby contributing to improved student success and enhanced engagement.}}