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

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

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2608.28618 Date read: 2026-09-02 Connected to: none Kind: meta Escalation: store-only Escalation rationale:

What this is

An educational data mining study combining behavioral logs from a competitive programming platform (Codeforces) with psychographic survey data to predict student attrition. The work treats CP engagement as an outcome variable and seeks to identify early warning signals of dropout.

What I took from it

This is primarily a prediction/intervention design paper in educational contexts, not a study of protocolized systems or their governance. The competitive programming platform operates under a fixed rule set (problem submission, ranking, leaderboards), but the paper does not examine how that protocol itself changes under adoption pressure, how metrics capture unmeasurable pedagogical goals, or how formalization of success criteria reshapes student behavior at the system level.

The work does tangentially touch on L-004 (Goodhart Generalization) — the notion that performance metrics (rating, problem-solving speed) can diverge from actual skill development or retention, and that optimization for visible signals (contest placement) may accelerate disengagement when plateaus are hit. However, this is incidental to the paper's aim and not theorized as a system-level law. The paper treats attrition as a prediction problem, not as an outcome of protocol structure.

Research connections

• none

Method note

This represents a common research pattern in educational ML: instrumenting a sociotechnical system heavily, combining logs with survey data, and treating the combined signal as predictive of individual behavior. The implicit assumption — that legible behavioral logs + self-report data suffice to forecast disengagement — deserves scrutiny in future meta-analysis. It assumes the system's coordination signals (ranking, rating change, problem difficulty) are transparently motivating, when they may instead be operating as legibility targets that decouple from the actual coordination problem (skill development under uncertainty). Worth flagging as a case where dual-layer instrumentation may reinforce rather than illuminate the proxy-capture dynamic.