Predictive Analytics for Student Retention & Fairness-Aware Intervention

📌 Project Overview

This project presents an end-to-end predictive analytics system designed to identify students at high risk of academic dropout using imbalanced educational datasets. Beyond predictive accuracy, the system integrates fairness-aware evaluation and cost–benefit analysis to ensure that data-driven retention strategies are both effective and equitable across demographic groups.

The project combines classical machine learning, dimensionality reduction, and fairness diagnostics to support early intervention decision-making in academic institutions.

---

🎯 Problem Statement

Student attrition poses significant academic and financial challenges for universities. Traditional predictive models often:

• Perform poorly on minority or at-risk populations
• Optimize accuracy without considering recall for early-risk identification
• Ignore demographic fairness and intervention costs

This project addresses these gaps by:

• Prioritizing recall-focused prediction of at-risk students
• Auditing algorithmic bias across demographic groups
• Evaluating the financial impact of predictive retention strategies

---

🧠 Key Objectives

• Predict student dropout risk on highly imbalanced datasets
• Maximize early-risk detection (recall-focused optimization)
• Reduce multicollinearity via dimensionality reduction
• Evaluate algorithmic fairness across demographic groups
• Quantify financial trade-offs of data-driven interventions

---

🛠️ Tech Stack

• Programming Language: Python
• Machine Learning: Scikit-learn
• Models: Random Forest, Support Vector Machine (SVM), Stacked Ensemble
• Dimensionality Reduction: Principal Component Analysis (PCA)
• Evaluation Metrics: Average Precision (AP), Recall, Confusion Matrix
• Fairness Analysis: False Positive Rate (FPR) disparity analysis
• Visualization: Matplotlib, Seaborn
• Environment: Jupyter Notebook

---

📂 Dataset Description

The project uses three structured datasets containing:

• Academic performance indicators
• Enrollment and progression attributes
• Demographic variables for fairness auditing

The target variable represents student persistence vs. dropout status.
All datasets exhibit class imbalance, motivating recall-focused modeling.

---

🔄 Project Pipeline

1️⃣ Data Ingestion & Exploration

• Loaded and inspected multiple academic datasets
• Assessed missing values, feature distributions, and class imbalance
• Identified potential data quality issues and inconsistencies

2️⃣ Data Preprocessing & Feature Engineering

• Cleaned and standardized numerical features
• Encoded categorical and demographic attributes
• Injected and analyzed synthetic data issues to test robustness
• Applied PCA to:
  • Reduce multicollinearity
  • Improve model stability across cross-validation folds

3️⃣ Modeling Strategy

• Trained baseline classifiers for comparison
• Built a stacked ensemble model combining:
  • Random Forest (nonlinear feature interactions)
  • Support Vector Machine (margin-based generalization)
• Optimized hyperparameters using cross-validation

4️⃣ Evaluation & Threshold Optimization

• Evaluated models using Average Precision (AP) due to class imbalance
• Achieved 0.93 Average Precision
• Tuned decision thresholds to maximize recall, improving early-risk detection compared to baseline classifiers

5️⃣ Fairness-Aware Model Audit

• Conducted subgroup analysis across demographic attributes
• Measured False Positive Rate (FPR) disparities
• Identified potential algorithmic bias in intervention targeting
• Compared fairness outcomes before and after threshold adjustments

6️⃣ Cost–Benefit Analysis

• Estimated financial impact of false positives vs. false negatives
• Assessed trade-offs between intervention costs and retention gains
• Demonstrated how recall-focused, fairness-aware models improve ROI for retention programs

---

📊 Key Results

• Average Precision: 0.93
• Improved Recall: Significant gains in early-risk student identification
• Stability: PCA reduced performance variance across CV folds
• Fairness Insights: Highlighted demographic disparities in false-positive rates
• Business Impact: Data-driven interventions showed favorable cost–benefit trade-offs

---

🔍 Why This Project Matters

This project goes beyond “just accuracy” by integrating:

• Imbalanced learning strategies
• Fairness diagnostics
• Financial reasoning

It demonstrates how machine learning can be responsibly applied in education to support equitable and effective decision-making.

---

▶️ How to Run

1. Clone the repository:

   https://github.com/atharvaa45/Predictive-Analytics-Student-Retention-Fairness-Aware-Intervention.git
   

2. Install dependencies:

   pip install numpy pandas scikit-learn matplotlib seaborn
   

3. Open the notebook:

   jupyter notebook
   

4. Run all cells sequentially in the main .ipynb file

🚀 Future Enhancements

Incorporate causal inference for intervention effectiveness

Extend fairness metrics (Equal Opportunity, Demographic Parity)

Deploy as an interactive dashboard for academic advisors

Automate model monitoring for drift detection

👤 Author

Atharva Honrao

Graduate Student — Data Science

Focus Areas: Machine Learning, Responsible AI, Applied Analytics