Student GPA Prediction Project

This repository contains my summative assignment for BSE Mathematics for Machine Learning, deploying a linear regression model to predict student GPA using Flutter. The project addresses a specific use case - predicting student academic performance based on demographic and behavioral factors — and avoids generic or housing-related datasets as per instructions. It is structured into four tasks:

• Task 1: Model training
• Task 2: API development
• Task 3: Flutter app creation
• Task 4: Video demo

File Structure

linear_regression_model/
├── summative/
│   ├── linear_regression/
│   │   ├── multivariate.ipynb  
│   ├── API/
│   │   ├── main.py             
│   │   ├── models.py           
│   │   ├── prediction.py       
│   │   ├── best_model.pkl      
│   │   ├── scaler.pkl          
│   │   ├── requirements.txt    
│   ├── FlutterApp/             
│   │   ├── lib/
│   │   │   ├── main.dart       
│   │   ├── pubspec.yaml        
├── README.md                   

Mission

Enhance student academic success by predicting GPA based on key factors, enabling targeted interventions for at-risk students.

Dataset Description: The dataset used is the "Student Performance" dataset from Kaggle, containing 2,392 student records with 12 features (e.g., Age, Gender, StudyTimeWeekly, Absences) and GPA as the target variable. It’s rich in volume (thousands of rows) and variety (demographic, academic, and extracurricular variables).

Sourced from: Kaggle: Students Performance Dataset

Visualizations:

1. Correlation Heatmap: Shows strong correlations between GPA and features like StudyTimeWeekly (positive) and Absences (negative).
2. GPA Distribution: Histogram of GPA reveals most students have GPAs between 2.5 and 3.5, reflecting typical academic performance and influencing model training

---

Task 1: Linear Regression Task

Description

• File: summative/linear_regression/multivariate.ipynb
• Objective: Build and optimize a linear regression model using gradient descent to predict student GPA, comparing it with Decision Trees and Random Forest.
• Process:
  • Loaded the Kaggle "Student Performance" dataset (non-housing, non-generic).
  • Preprocessed data (normalized using StandardScaler).
  • Trained three models using scikit-learn:
    1. Linear Regression
    2. Decision Trees
    3. Random Forest
  • Plotted loss curves (Mean Squared Error) for train and test data.
  • Compared performance:
    • Linear Regression: MSE = 0.15 (lowest loss)
    • Random Forest: MSE = 0.18
    • Decision Trees: MSE = 0.22
  • Saved the best model (Linear Regression) as best_model.pkl and scaler as scaler.pkl.
  • Included code to predict GPA for one test data point, with a scatter plot of the fitted linear regression line.

Artifacts

• best_model.pkl: Trained Linear Regression model.
• scaler.pkl: Scaler for input normalization.

---

Task 2: API Development

Description

• Directory: summative/API/
• Objective: Create a FastAPI endpoint to predict GPA using the Task 1 model.
• Endpoint: /predict (POST)
• Input: JSON with 12 fields (e.g., Age, Gender, StudyTimeWeekly).
• Output: JSON with predicted_gpa, grade_class (A-F), and grade_value (0-4).

Files

• main.py: FastAPI app with CORS middleware.
• models.py: Pydantic BaseModel enforcing data types (e.g., int, float) and ranges (e.g., Age: 15-18, StudyTimeWeekly: 0-20).
• prediction.py: Loads model/scaler, predicts GPA, and maps to grade class.
• requirements.txt: Lists dependencies (fastapi, pydantic, uvicorn, scikit-learn, joblib).

Deployment

• Platform: Render
• Public URL: https://student-performance-api-wknc.onrender.com/docs
• API Endpoint: https://student-performance-api-wknc.onrender.com/predict
• Instructions:
  1. Pushed summative/API/ to GitHub.
  2. Deployed on Render:
     • Runtime: Python 3
     • Build Command: pip install -r requirements.txt
     • Start Command: uvicorn main:app --host 0.0.0.0 --port $PORT
  3. Verified Swagger UI at /docs.

Example Request

{
  "Age": 16,
  "Gender": 0,
  "Ethnicity": 0,
  "ParentalEducation": 2,
  "StudyTimeWeekly": 10.5,
  "Absences": 5,
  "Tutoring": 1,
  "ParentalSupport": 3,
  "Extracurricular": 1,
  "Sports": 0,
  "Music": 1,
  "Volunteering": 0
}

Response

{
  "predicted_gpa": 3.75,
  "grade_class": "A",
  "grade_value": 0
}

---

Task 3: Flutter App

Description

• Directory: summative/FlutterApp/
• Objective: Build a mobile app to interact with the Task 2 API.
• Features:
  • Two pages: Home (welcome screen) and Prediction.
  • 12 input fields (10 dropdowns, 2 text fields for StudyTimeWeekly and Absences).
  • "Predict" button.
  • Output display showing GPA and grade, or error messages.

Running Locally

1. Navigate to summative/FlutterApp/.
2. Install dependencies:

   flutter pub get

3. Run the app:

   flutter run

Task 4: Video Demo

Description

• Link: Demo Video URL
• Duration: ~5 minutes

---