Temperament as Fate? Predicting Academic Risk through Educational Data Mining

TL;DR

Can a student's personality predict their likelihood of failing out of an IT program? Researchers at HSE-Nizhny Novgorod utilized Educational Data Mining (EDM) to prove that psychological temperament—ranging from "hot-tempered" Cholerics to "sensitive" Melancholics—significantly impacts academic success. By applying Decision Tree models, they reached an 84% accuracy in identifying "risky" students before they officially failed their exams.

Background: The 30% Dropout Dilemma

In highly selective IT programs, entry scores are rarely the problem. Students arrive with high Unified State Exam results, yet nearly 30% vanish after the first year. This research identifies a critical "blind spot": the social and psychological isolation of IT students (often introverts) who struggle to navigate the high-pressure environment of Mathematical Analysis and Linear Algebra.

The Core Insight: Mapping Mind to Mark

The authors moved beyond simple grade tracking. They hypothesized that temperament provides the foundational "Inductive Bias" for how a student handles stress and workload. Using the Eysenck Personality Questionnaire, they mapped students across two axes: Extraversion-Introversion and Stability-Instability.

The Four-Quadrant Personality Framework:

• Choleric: Energy-rich but potentially unstable; often show lower GPAs due to impulsivity.
• Sanguine: Hard-working and cheerful; lowest rate of retaken exams.
• Phlegmatic: Slow but persistent; high perseverance leads to higher marks.
• Melancholic: Deeply emotional and sensitive; highest risk of academic failure.

Methodology: Decision Trees vs. kNN

The researchers compared two primary machine learning architectures to classify students into three risk tiers: Good (Safe), Medium (At Margin), and Bad (High Risk).

• Stage 1: Feature Selection: Correlations revealed that "Perseverance" (0.19) and "Ability to Prioritize" (0.29) were the strongest predictors.
• Stage 2: Model Training: While kNN (k-Nearest Neighbors) reached 77% accuracy, it struggled with "noisy" psychological data.
• Stage 3: The Winner: The Decision Tree model achieved 84% accuracy by prioritizing specific trait combinations (e.g., Lack of Perseverance + Melancholic Temperament = High Risk).

Experimental Results & Validation

The model was put to the ultimate test: predicting the outcomes of 40 first-year students before the summer session.

The results were striking:

• 64% of predicted "Bad" students faced actual retakes.
• 27% of that group dropped out or were placed on Individual Learning Plans.
• 0% of the predicted "Good" students had any academic debt.

Critical Insight: Why This Matters

The study concludes that "hot-tempered" Cholerics and "sensitive" Melancholics are structurally disadvantaged in traditional IT curricula. Cholerics may crash due to high-intensity bursts followed by burnout, while Melancholics may succumb to the pressure of failure (the "retake loop").

Recommendations for Educational Offices:

1. Early Warning Systems: Implement surveys in October to flag "risky" profiles.
2. Targeted Support: Provide additional tutoring exclusively for the "Bad" and "Medium" risk categories.
3. Human-in-the-Loop: Educational offices should conduct personal interviews with predicted "Bad" category students before the first exam session.

Limitations & Future Work

The sample size (140 students) is relatively small for "Big Data." However, the high precision of the Decision Tree suggests that even with "Small Data," psychological features provide a powerful signal. Future research directions include applying Process Mining to see how these temperaments navigate the Learning Management System (LMS) in real-time.

Takeaway

Success in STEM isn't just about IQ; it's about the interaction between one's nervous system and the curriculum. By quantifying temperament, universities can transform from reactive "testing centers" into proactive "support ecosystems."