# What is Data Leakage? Definition & Technical Mechanics

> **Source:** [SPIDITS AI AI Glossary](https://spidits.com/ai-glossary)  
> **Category:** Model Training  
> **Canonical Citation:** [https://spidits.com/ai-glossary/data-leakage](https://spidits.com/ai-glossary/data-leakage)

## Definition
Data Leakage is a training error that occurs when information from outside the training dataset is used to train a model. This leads to overly optimistic performance scores during validation, but poor generalization on true unseen data.

## Architectural Mechanics & Deep Dive
Data leakage occurs when information from outside the training dataset is accidentally used to train a machine learning model, leading to overly optimistic performance estimates during training and validation that fail to replicate on real-world data. Common sources of leakage include pre-processing features across the entire dataset before splitting it into train/test sets, or including features that target information that would not actually be available at the time of prediction.

## Industry Applications & Real-World Use Cases
Dataset splitting audits, feature engineering checks, and cross-validation pipelines.

## Frequently Asked Questions

### How does data leakage typically happen?
By scaling or normalizing the entire dataset (including training and validation sets) together, rather than calculating scaling parameters only on the training split.

### How do you prevent data leakage?
Strictly separate the validation and test datasets before performing any preprocessing, cleaning, or feature engineering transformations.

## Related Terminology
- [validation-data](https://spidits.com/ai-glossary/validation-data)
- [overfitting](https://spidits.com/ai-glossary/overfitting)
- [bias-variance-tradeoff](https://spidits.com/ai-glossary/bias-variance-tradeoff)

---
*From the SPIDITS AI Knowledge Base (336+ terms indexed). Explore full technical definitions: [https://spidits.com/ai-glossary](https://spidits.com/ai-glossary)*