Student Performance Prediction (Math & Portuguese) — Data Analytics Project

Executive Summary

This project analyzes secondary school student achievement in two Portuguese schools. Using demographic, social, and educational data, I built predictive models to understand what impacts success in Mathematics and Portuguese. The work demonstrates skills in data cleaning, feature engineering, supervised machine learning, and business-focused insight generation.

---

Problem

Student performance is influenced by a wide range of factors — from attendance and paid classes to aspirations for higher education. The challenge was to:

1. Predict the first-period Math grade (G1.Math) without using other grade features.
2. Classify G1.Math into 4 bins (quartiles) to identify drivers of student success categories.
3. Respect strict constraints: include variables (paid.Math, higher, absences.Port, paid.Port) and exclude (Fedu, address, Mjob).

---

Methodology

• Data Preparation: Cleaned data, encoded categorical variables, handled missing values, and enforced feature constraints.
• Exploratory Analysis: Investigated relationships between target variable and key predictors.
• Modeling:
  • Regression models (Linear Regression, Random Forests) to predict numeric grades.
  • Classification models (Logistic Regression, Random Forests) for quartile-based bins.
• Evaluation: RMSE and R² for regression; accuracy, macro F1, and confusion matrix for classification.
• Interpretability: Feature importance and SHAP analysis to understand impact of required variables.

---

Skills Demonstrated

• Data wrangling with Python (pandas, numpy)
• Statistical analysis & EDA (matplotlib, seaborn)
• Predictive modeling with scikit-learn
• Classification & regression evaluation
• Feature engineering & categorical encoding
• Communicating insights with visuals and structured reporting

---

Results & Business Recommendations

• Key Drivers:
  • Students who attend paid Math classes consistently perform better.
  • Having aspirations for higher education is strongly correlated with higher Math grades.
  • Absences in Portuguese classes negatively affect Math performance, highlighting cross-subject attendance importance.
• Regression Performance: Achieved strong predictive accuracy (R² ~0.58, RMSE ~0.68).
• Classification Performance: Quartile model achieved accuracy of ~68% with balanced class distribution.

Recommendations:

• Expand access to affordable tutoring/paid classes for students from lower-income backgrounds.
• Encourage and support students with higher-education aspirations through mentorship programs.
• Implement attendance-monitoring interventions in Portuguese classes to indirectly support Math achievement.

---

Next Steps

• Extend analysis to longitudinal performance trends across subjects.
• Explore causal relationships (e.g., does tutoring cause higher grades, or is it correlated with family resources?).
• Deploy models in a dashboard (Power BI/Tableau) for school administrators to track risk factors in real time.
• Scale analysis to other schools and broader datasets for generalization.