Predicting Student Attrition in Competitive Programming: A Large-Scale Study Integrating Survey Insights and Global Behavioral Logs

A dual-layer predictive framework for understanding student attrition in competitive programming, combining large-scale Codeforces behavioral logs (n = 1,816 after filtering and balancing) with a multi-institutional psychographic survey from 10 universities in Bangladesh (n = 64 for predictive modeling).

Behavioral findings: what precedes attrition

The behavioral analysis confirmed that true attrition is preceded by an 83.71% reduction in contest participation and consistent underperformance on skill-related metrics. The authors also validate a Skill-Application Paradox: students who stop report higher theoretical confidence than their active peers yet exhibit significantly weaker practice habits (significant at p < .001 for upsolving habit and peer circle density). This disconnect between self-reported confidence and actual practice behavior is a distinctive Self Efficacy signal for early-warning.

Predictive modeling

Machine learning benchmarks demonstrated that a Soft-Voting Ensemble achieved the strongest performance on the behavioral dataset (CV F1 = 0.737, Test Recall = 0.769), while Random Forest led on the survey dataset (CV F1 = 0.924, interpreted as a localized exploratory pilot). An engineered Intensity Ratio ranked as the third most predictive Codeforces feature, capturing independent practice effort beyond raw activity counts.

Early Warning System proof of concept

Applied as a proof-of-concept Early Warning System, the survey-trained model identified four high-risk active students whose behavioral profiles corroborated the model's predictions. This demonstrates the feasibility of flagging at-risk students from mixed survey and behavioral signals.

Connected Concepts

• CS Education
• Learning Analytics
• Self Efficacy
• Motivation
• Student Engagement
• Assessment
• Personalized Learning
• Student Modeling
• Self Report Measures

Connected Articles

• AI Writes Code Student Writes Model 2026 — AI Writes Code, Student Writes Model
• AI Literacy Equity Programming Policy — AI Literacy, Equity and Programming Policy
• Virtual Tutoring Computer Assisted Learning Takeup 2026 — Virtual Tutoring and Computer-Assisted Learning Take-Up
• Affective Text Wearable Student Health — Affective Text and Wearable Student Health

Citation

Ruhan, A. I., Naeem, G. M., Rafi, R. I., Mim, S. A., Opi, N. B., Chowdhury, D. F., & Sadi, M. R. K. (2026). Predicting Student Attrition in Competitive Programming: A Large-Scale Study Integrating Survey Insights and Global Behavioral Logs. arXiv:2608.28618.