IELTS Score Predictor ML

A Machine Learning pipeline for automated IELTS writing scoring, predicting band scores across four rubric categories.

Academic Context: Final Exam for the Artificial Intelligence course (2025). Ranked via a Kaggle competition.

---

🏗 Architecture & Methodology

This project implements a regression pipeline to estimate IELTS band scores (0-9) based on essay content. It handles multiple target variables simultaneously.

Key Steps:

1. Preprocessing (src/preprocess.py):
   • Data Cleaning: Removes punctuation/numbers and strips excess whitespace using Regex.
   • Feature Engineering: Concatenates prompt and essay to provide full context to the model.
   • Vectorization: Uses TF-IDF (Term Frequency-Inverse Document Frequency) to convert text into numerical features (max 10,000 features).
2. Model Training (src/train.py):
   • Algorithm: MultiOutput Regressor wrapping a Random Forest (100 estimators).
   • Why MultiOutput? IELTS essays are graded on 4 distinct criteria simultaneously. This architecture predicts all 4 scores in a single pass.
3. Evaluation:
   • Uses Mean Squared Error (MSE) on a validation split (20%) to assess prediction accuracy.

---

📂 Project Structure

├── data/
│   ├── df_train.csv       # Training data (Prompt + Essay + 4 Scores)
│   └── df_test.csv        # Test data (Prompt + Essay only)
├── notebooks/
│   └── eda.ipynb          # Exploratory Data Analysis
├── src/
│   ├── preprocess.py      # Cleaning and TF-IDF vectorization logic
│   ├── train.py           # Random Forest training & evaluation functions
│   └── predict.py         # Submission file generation
├── outputs/
│   └── submission.csv     # Generated predictions
├── main.py                # Pipeline orchestrator
└── requirements.txt       # Dependencies

---

🚀 Getting Started

Prerequisites

• Python 3.8+
• Packages: pandas, scikit-learn

Installation

git clone [https://github.com/noecrn/IELTS-Score-Predictor-ML.git](https://github.com/noecrn/IELTS-Score-Predictor-ML.git)
cd IELTS-Score-Predictor-ML
pip install -r requirements.txt

Usage

Run the complete training and prediction pipeline:

python main.py

This script will:

1. Load and clean df_train.csv.
2. Train the Multi-Output Random Forest model.
3. Print the Validation MSE.
4. Generate score predictions for df_test.csv in outputs/submission.csv.

---

📊 Targets & Metrics

The model predicts scores for the official IELTS writing rubric:

• Task Achievement
• Coherence and Cohesion
• Lexical Resource
• Grammatical Range

Metric: Mean Squared Error (MSE).

---

👥 Authors

• Noé Cornu - Engineering Student @ EPITA - GitHub | LinkedIn
• Baptiste Rio - Engineering Student @ EPITA