hmnshudhmn24/dataset-quality-lm

Dataset Quality LM 📊🧪

DatasetQualityLM is an AI system that evaluates datasets for bias, data leakage, noise, and real-world deployability before model training.

It helps teams detect hidden dataset risks early, improving model reliability, fairness, and production readiness.

---

🔍 What Problem Does It Solve?

Many ML failures come from bad data, not bad models.

DatasetQualityLM answers:

• Is this dataset biased?
• Does it contain data leakage?
• How noisy or inconsistent is it?
• Is it safe to deploy models trained on it?

---

✨ Key Features

• ⚖️ Bias detection (demographic & distributional)
• 🔓 Target & feature leakage detection
• 🔊 Noise and missing-value analysis
• 🚀 Deployability scoring (single quality score)
• 🧠 Explainable, rule-based analysis
• 🤗 Hugging Face–ready pipeline
• 🎛️ Gradio demo included
• 🧪 Unit-tested core components

---

📂 Project Structure

dataset-quality-lm/
├── config/
├── data/
├── src/
├── training/
├── pipelines/
├── scripts/
├── tests/
├── notebooks/
├── app.py
├── README.md
├── model_card.md
├── requirements.txt
└── LICENSE

---

⚙️ Installation

pip install -r requirements.txt

---

🚀 Quick Usage

from src.inference import DatasetQualityPipeline

pipeline = DatasetQualityPipeline()

result = pipeline("data/samples/clean_dataset.json")
print(result)

---

🎛️ Gradio Demo

python app.py

---

🧠 How It Works

1. Dataset Loading
2. Bias Detection
3. Leakage Detection
4. Noise Analysis
5. Deployability Scoring