Student Performance — Model Comparison
A premium, serverless Machine Learning Studio that compares 7 regression models with full EDA, dynamic leaderboard plots, and in-browser predictions using exported coefficients.
About the Project
The Student Performance Model Comparison Studio is a full-featured ML dashboard designed with a custom Sunset Warm theme (amber, orange, and alabaster cream). It compares 7 regression models (Linear Regression, Ridge, Lasso, Decision Tree, Random Forest, Gradient Boosting, and SVR) on a dataset of 1,000 students to analyze how demographics, preparation courses, and lunch types affect academic achievement.
Model Training & Inference Pipeline
To bypass cross-origin resource sharing (CORS) blocks and support serverless client-side evaluation, this project operates on a custom end-to-end ML pipeline:
Dataset
1,000 Records
Prep
One-hot Encoding
Training
7 Regressors
Validation
5-Fold CV
Leaderboard
Rank & Compare
Key Features
Packed with interactive diagnostic tools and analytics panels:
Sunset Predictor: Slider form with conic-gradient score gauges and dynamic performance feedback.
Exploratory Data Analysis: Real-time histograms for score distributions, inter-feature correlation heatmaps, and category box plots.
Model Leaderboard: Ranks algorithms based on average R² performance, with an overlay radar chart comparing the top 3 models.
Actual vs. Predicted Plot: Scatter comparison charts mapping model performance relative to a perfect prediction line (y = x).
Zero Network Overhead: Chart.js is bundled locally (chart.umd.min.js) to support 100% offline execution.
Tech Stack
Built using a combination of lightweight client scripts and Python ML tooling:
Frontend: HTML5, Vanilla CSS3 (custom CSS variables & gradients), JavaScript ES6
Visualization: Chart.js (locally bundled v4.x library)
ML Engine (Training): Python 3.11, Scikit-learn (7 regression algorithms, Standard Scaler, Label Encoder)
Data Wrangling: Pandas, NumPy
Python Alternative: Streamlit, Plotly, Seaborn, Matplotlib
Project Demo
Project Note
🔌 Dual Architecture Support This studio features two implementations: 1. Static Web Studio (Client-Side): A fast, lightweight, and fully offline-compatible dashboard in HTML/CSS/JS using Chart.js that computes predictions instantly inside the browser. 2. Streamlit Cloud (Server-Side): A Python-based analytics server utilizing Streamlit, Plotly, Seaborn, and Matplotlib. Both projects are housed within the unified GitHub repository.