AI Can Now Predict Who Will Quit Coding — Before They Give Up

Competitive programming is a great way for students to sharpen problem-solving skills, but many quietly give up when they hit a plateau or feel anxious. Until now, no one had studied exactly why they leave. Researchers in Bangladesh looked at activity logs from Codeforces (a popular coding-contest website) and surveyed university students to find the earliest warning signs.

The results reveal a surprising "Skill-Application Paradox." Students who quit actually rated their math and data-structures knowledge higher than students who stayed — but they did far less independent practice and rarely reviewed their contest mistakes. On behavior, quitting was almost always preceded by a dramatic 83.7% drop in contest participation and a 15.6% increase in time spent on difficult problems. In other words, they don't quit because they feel dumb; they quit even though they think they're good.

The team built an AI early-warning system that combines behavioral data (how often you compete, how long you struggle) with a simple survey. The survey-based model was especially strong, correctly identifying at-risk students about 92% of the time. In a pilot run with 22 active students, the system flagged 4 as high risk — all of whom, the authors say, displayed the pattern before any human mentor noticed.

What does this mean for everyday people? For universities and online learning platforms, it's a way to provide just-in-time help: a nudge to practice, a mentor check-in, or a study group invite before a student disappears. It's also a reminder that confidence alone doesn't equal progress — real improvement comes from consistent, humble practice. The system is still a proof of concept on a small scale, but it points toward smarter, more empathetic education tech.