ID3

ID3, short for Iterative Dichotomiser 3, is a decision tree induction algorithm developed by Ross Quinlan as part of his work on decision tree learning in the 1980s. It is used for supervised classification by constructing a decision tree from labeled examples. The algorithm selects at each step the attribute that yields the largest information gain with respect to the target class, measured using entropy. Each chosen attribute splits the training data into subsets corresponding to its values, and the process recurses on the subsets until they are homogeneous or no remaining attributes exist. ID3 assumes categorical attributes and nominal classes, and it can overfit with noisy data unless pruning is applied. It is relatively simple and fast, making it a foundational method in early machine learning pedagogy.

ID3 contributed to later developments such as C4.5, which generalized the approach to handle continuous attributes through discretization or thresholding, supports pruning to reduce overfitting, handles missing values, and produces more compact trees. While modern learners may favor ensemble methods, ID3 remains a common teaching example of information-gain-based tree induction.

ID3 can also refer to ID3 tags, a metadata container used in MP3 audio files. Versions ID3v1 and ID3v2 store information like title, artist, album, and track duration; ID3v2 adds more flexible framing and extended features and is widely supported by many media players.