strova-ai/credit_scoring_datatset

🏦 Synthetic Credit Scoring Dataset — Powered by Syncora

🌐 Official Website: Syncora.ai

High-fidelity synthetic financial behavior dataset for AI, ML modeling & LLM training.

---

Dataset Summary

This dataset contains synthetic financial records simulating customer behavior in a credit scoring context.
Generated with Syncora.ai, it provides privacy-safe, realistic data while preserving statistical fidelity.

Key applications:

• Credit risk modeling
• Machine learning classification
• Feature engineering for financial AI
• Dataset for LLM training (tabular-to-text, reasoning with structured finance data)
• Educational use in data science courses

---

📊 Dataset Info

Field	Details
Features	- CUST_ID (string) - INCOME (int32) - SAVINGS (int32) - DEBT (int32) - CREDIT_SCORE (int32) - DEFAULT (int32)
Task Categories	- Tabular Classification - Financial Risk Modeling
License	Apache-2.0
Size Category	10K < n < 100K

Format: CSV, ~20K synthetic records.

---

📦 What This Repo Contains

• Synthetic Credit Scoring Dataset – CSV format, ready for ML modeling.
  ⬇️ Download Dataset

• Jupyter Notebook – Exploration and usage guide for the dataset.
  📓 Open Notebook

• Syncora Platform – Generate your own high-fidelity synthetic datasets.
  ⚡ Generate Your Own Synthetic Data

🤖 Machine Learning & AI Use Cases

• 💳 Credit Risk Modeling: Train classification models to predict default risk.
• ⚙️ Feature Engineering: Extract behavioral features like debt-to-income and repayment consistency.
• 🧠 LLM Alignment: Use as a structured dataset for LLM training (e.g., converting tabular inputs into human-readable risk assessments).
• 📊 Benchmarking: Compare model accuracy, precision, and recall across logistic regression, random forest, XGBoost, and deep learning.
• 🔍 Explainability: Apply SHAP, LIME, or ELI5 to interpret model predictions.
• ⚖️ Bias & Fairness Studies: Explore whether synthetic datasets can reduce bias compared to real-world financial data.
• ✅ Synthetic Data Validation: Test how well synthetic datasets maintain model performance relative to real datasets.

Usage

Load directly with Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("syncora-ai/synthetic-credit-scoring")
print(dataset["train"][0])