Labeled data

Labeled data is raw data annotated with meaningful tags (labels), providing the crucial ground truth required for training accurate, production-grade supervised machine learning models.

Labeled data is the absolute foundation for supervised ML: it's raw input (images, text, audio) meticulously tagged by human annotators to create the 'answer key' a model learns from. Consider image recognition: a human draws a bounding box and assigns the label 'car' or 'person' to specific pixels. This annotation process is resource-intensive, but it's non-negotiable for building high-performing models in computer vision, NLP (e.g., sentiment analysis), and speech recognition. The quality of this labeled dataset directly dictates the accuracy of the final model's predictions on new, unlabeled data.

https://cloud.google.com/learn/what-is-data-labeling