Data Labelling is the process of tagging raw data with the correct answers or categories, creating the labelled examples that supervised machine learning models learn from.
What it means in plain English
Supervised models learn by example, and those examples need correct answers attached. Data labelling is the work of adding them — marking which emails are spam, drawing boxes around objects in images, or transcribing audio. It is often done by people, and it is one of the most time-consuming and important parts of building a model, because the quality of the labels directly shapes the quality of the model.
Poor or inconsistent labels lead to poor models, no matter how good the algorithm.
A simple example
To build a model that detects defective products from photos, workers go through thousands of images labelling each as “defective” or “good.” That labelled set becomes the training data the model learns from.
Why it matters
Data labelling is the often-unseen foundation of supervised AI. Its quality determines how well a model performs, which is why so much effort — and an entire industry — is devoted to producing accurate, consistent labels.
Related terms
- Training Data — what labelling produces.
- Supervised Learning — depends on labelled data.
- Ground Truth — the correct labels a model is judged against.
Frequently asked questions
What is data labelling?
It is the process of annotating data with the correct answers (labels) — such as tagging images with what they contain — so a supervised model can learn from them.
Why is data labelling important?
The quality of labels directly affects model quality: inaccurate or inconsistent labels lead to a worse model, which is why labelling is often careful, expensive, and sometimes expert work.
