🎯 AutoJudge – Problem Difficulty Predictor

AutoJudge is an end-to-end machine learning application that predicts the difficulty level of programming problems based on their textual content.
The system analyzes problem titles and descriptions using NLP techniques and outputs a difficulty classification through an interactive Streamlit web app.

This project demonstrates a complete ML workflow — from preprocessing and feature engineering to model selection and deployment.

---

📌 Problem Statement

Online coding platforms host thousands of problems with varying difficulty levels.
Manual difficulty tagging is subjective, inconsistent, and time-consuming.

Goal:
To automatically predict the difficulty level of a problem using its title and description, ensuring consistency and scalability.

---

🧠 Approach Overview

The project follows a structured machine learning pipeline:

1. Text preprocessing
2. Feature extraction (TF-IDF + engineered features)
3. Model training & evaluation
4. Best model selection
5. Deployment using Streamlit

---

🧾 Input Features

The model takes the following inputs:

• Problem Title (text)
• Problem Description (text)

Engineered Features

• TF-IDF vectors from title and description
• Text length and word-count based numerical features
• Combined sparse and dense features using hstack

---

🛠️ Text Preprocessing

Text data is cleaned and normalized using:

• Lowercasing
• Regex-based cleaning
• Whitespace normalization
• Removal of unwanted characters

Preprocessing is identical during training and inference to avoid data leakage.

---

⚙️ Feature Engineering

• TF-IDF vectorization for:
  • Problem titles
  • Problem descriptions
• Numeric features:
  • Character count
  • Word count
• Final feature matrix created by combining:
  • Sparse TF-IDF features
  • Dense numerical features

---

🤖 Models Trained

Multiple models were trained and evaluated, including:

• Logistic Regression (baseline)
• Tree-based models
• Gradient Boosting–based models

Each model was evaluated using appropriate performance metrics, and the best-performing model was selected for deployment.

---

🏆 Final Model

• Best model selected based on evaluation performance
• Model and vectorizers serialized using pickle
• Loaded safely during inference in the Streamlit app

---

• 📂 Project Files (Google Drive)

🌐 Web Application (Streamlit)

The project includes an interactive Streamlit web app that:

• Accepts user input (title + description)
• Applies the same preprocessing and feature extraction
• Loads the trained model and vectorizers
• Outputs the predicted difficulty level

---

🗂️ Project Structure

AutoJudge/
│
├── model/
│   ├── scaler.pkl
│   ├── vectoriser.pkl
│   ├── label_encoder.pkl
│   ├── classifier.pkl
│   ├── regressor.pkl
│   └── svd.pkl
│
├── difficulty_predictor.ipynb
├── app.py
├── autojudge.pdf
├── requirements.txt
└── README.md