🧠 AI Writing Style Fingerprinter

⚠️ Note: This project was tested with a small, random dataset of ~30–40 examples, mostly for experimentation.
Feel free to create your own dataset to train a more accurate and robust model.

---

📌 Overview

AI Writing Style Fingerprinter is a lightweight machine learning project designed to detect whether a paragraph was written by ChatGPT, Claude, or a Human, based purely on shallow linguistic features.

It uses:

• Part-of-speech ratios
• Sentence structure statistics
• Punctuation and stopword frequency

All without deep learning—only spaCy, scikit-learn, and some classic ML.

---

📂 Dataset Format

The dataset should be a CSV file named dataset.csv with the following structure:

text,label
"AI helps personalize education.",ChatGPT
"The cat sat on the mat.",Human
"By leveraging ML, performance improves.",Claude

---

⚙️ Features Extracted

• Average sentence length
• Punctuation count
• Stopword ratio
• Noun / Verb / Adjective ratios (POS-based)
• Number of sentences and tokens

---

🚀 Getting Started

1. Install dependencies

pip install pandas scikit-learn spacy matplotlib
python -m spacy download en_core_web_sm

2. Run the model

open  main.ipynb and run the code

This will:

• Load the dataset
• Extract linguistic features
• Train a RandomForestClassifier
• Print a classification report and confusion matrix
• Allow optional predictions on custom text

---

🧪 Example Usage

predict("AI can enhance teaching by providing real-time feedback.")
# ➜ Prediction: ChatGPT

---

📈 Outputs

• Confusion matrix plot (matplotlib)
• Classification metrics: Precision, Recall, F1-score
• Feature importance bar chart

---

📌 Notes

• Designed for quick testing; feel free to expand dataset for better performance.
• You can integrate Streamlit to create a live interface if needed.

---

🧠 Author

Pujan Neupane GitHub • Portfolio