Using Machine Learning to Predict and Prevent Academic Failure Before It Happens.

This project presents an AI-powered Early Warning System designed to identify students at risk of academic failure using machine learning models. By analyzing a range of academicbehavioraland social featuresthe system aims to predict whether a student will pass or fail — enabling timely interventions by educators.

To build a predictive system using supervised machine learning models that:

• Forecasts student academic success based on historical data
• Identifies key features contributing to student performance
• Enables early intervention to prevent failure

• Source: UCI Machine Learning Repository - Student Performance Dataset
• Focus: student-mat.csv (Math subject data only)
• Size: 395 records33 features

• G3 - Final grade
• Transformed into binary pass (1 if G3 >= 10 else 0)

• Demographic: sexageaddressfamsizeetc.
• Academic: studytimefailuresabsencesetc.
• Social: romanticgoouthealthetc.

• No missing values found in the dataset.

• Used Label Encoding on categorical columns such as schoolsexaddressetc.

• Applied StandardScaler to all numeric features except the target to normalize distributions and improve model performance.

• Dropped intermediate grade columns G1 and G2 which were highly correlated with G3 to ensure realistic evaluation.

• Visualized distribution of the pass label.
• Used heatmaps to inspect feature correlations.
• Found high correlation between past grades and final result.
• Observed socio-academic behavior patterns in successful students.

Used GridSearchCV for hyperparameter tuning and selected models based on accuracy and performance metrics:

• Logistic Regression
• Support Vector Classifier (SVC)
• Decision Tree
• Random Forest
• Gradient Boosting
• XGBoost (optional if system supports)

• Accuracy Score
• Classification Report
• Confusion Matrix

• Model: Random Forest / XGBoost (based on system performance)
• Accuracy Achieved: ~96%
• Key Insight: Study timefailuresabsencesand social factors like going out and romantic relationship status significantly impact academic performance.

• Schools/Colleges: Alert system for teachers to flag at-risk students
• EdTech Startups: Adaptive learning platforms to provide remedial content
• Policymakers: Tailored support strategies based on data-driven insights

1. Clone the repository
2. Install dependencies using requirements.txt
3. Run the Jupyter Notebook student_performance.ipynb
4. Review final model performance and visual insights

AI_Early_Warning_Student_Risk/
├── student-mat.csv
├── student_performance.ipynb
├── README.md
└── requirements.txt

• Python
• PandasNumPyMatplotlibSeaborn
• scikit-learn
• XGBoost (optional)

Suyog Manke Data Science & AI Enthusiast Passionate about solving real-world problems using ML and AI.

If you're using a Mac and facing issues with XGBoostensure OpenMP is installed:

brew install libomp

👨‍💻 Author Suyog Manke Powered with ❤️ by Suyog Manke

LinkedIn | GitHub | Email

“The goal is not just prediction — it’s timely intervention that changes outcomes.”