📊 Student Exam Score Prediction

This project was inspired by Chapter 2 of Aurélien Géron's Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow, where he performs detailed analysis on the Housing dataset. Motivated by that, I explored and applied similar machine learning techniques on the Student Habits and Academic Performance dataset to predict exam scores.

📁 Dataset

• Source: Kaggle - Student Habits and Academic Performance Dataset
• Description: Includes student habits, demographics, and academic indicators.
• Target: exam_score

🚀 Objective

Build and evaluate a regression model/Random Forest Model that can accurately predict a student’s exam score based on their habits and academic traits.

🧠 Features Included

• student_id: Unique identifier for each student.
• age: Age of the student (ranging from 16 to 28 years).
• gender: Gender identity (Male, Female, or Other).
• major: Student’s field of study (e.g., Computer Science, Engineering, Arts).
• study_hours_per_day: Average number of hours spent studying daily.
• social_media_hours: Daily time spent on social media platforms.
• netflix_hours: Daily time spent watching Netflix or similar streaming services.
• screen_time: Total daily screen time across devices.
• part_time_job: Indicates if the student holds a part-time job (Yes/No).
• attendance_percentage: Academic attendance rate as a percentage.
• sleep_hours: Average hours of sleep per night.
• exercise_frequency: Frequency of physical exercise (e.g., times per week).
• diet_quality: Qualitative rating of diet (e.g., poor, average, good).
• mental_health_rating: Self-assessed mental health on a scale from 1 to 10.
• stress_level: Perceived stress level on a scale from 1 to 10.
• exam_anxiety_score: Anxiety level related to exams on a scale from 1 to 10.
• extracurricular_participation: Involvement in extracurricular activities (Yes/No).
• access_to_tutoring: Whether the student has access to tutoring support (Yes/No).
• family_income_range: Family income category (e.g., low, medium, high).
• parental_support_level: Level of parental support (qualitative or numeric scale).
• parental_education_level: Highest education level attained by parents.
• motivation_level: Self-reported motivation on a scale from 1 to 10.
• time_management_score: Effectiveness in managing time, rated 1 to 10.
• learning_style: Preferred learning method (e.g., visual, auditory, kinesthetic).
• study_environment: Typical study location (e.g., home, library, café).
• dropout_risk: Binary indicator (Yes/No) derived from stress and motivation levels.
• previous_gpa: GPA from previous semesters or years.
• exam_score: Target variable — student’s exam score to be predicted.

🧪 Pipeline Steps

• Data Exploration and Cleaning
• Feature Engineering
• Pipelines with Custom Transformers
• Model Training (Linear Regression, Random Forest, etc.)
• Hyperparameter Tuning (GridSearchCV)
• Evaluation using RMSE and Error Percentage
• Visualizations using Seaborn & Matplotlib

💡 Results

• Best RMSE achieved: 4.180963504867681
• Error % ranged between: 0% - 4%

📈 Key Visualizations

• Scatter plots with color-coded percentage error
• Distribution of predicted vs actual scores
• Correlation heatmaps

Visualizations

Here are some key plots from the project:

Distribution of CGPA among students.

Comparison of predicted exam scores vs actual scores.

Distribution of Root Mean Squared Errors across predictions.

Relationship between study time and exam scores.

📚 Tools Used

• Python (Pandas, NumPy, Scikit-learn)
• Jupyter Notebook
• Seaborn, Matplotlib
• GridSearchCV for hyperparameter tuning

🗂️ Project Structure

├── student_exam_prediction.ipynb ├── data/ │ └── student_data.csv ├── images/ │ └── visualizations.png ├── requirements.txt └── README.md

📬 Contact

For questions or suggestions, feel free to reach out: Akshat Khatri 📧 [email protected]