Monilabelluokitus

Monilabelluokitus, or single-label classification, is a supervised learning task in which each input instance is assigned exactly one label from a predefined set of mutually exclusive classes. The goal is to learn a mapping from feature vectors to discrete categories that generalizes to unseen data. This contrasts with multi-label classification, where an instance may belong to several classes at once.

Typical methods include logistic regression, support vector machines with a multinomial or one-vs-rest approach, decision trees, random forests, Naive Bayes, k-nearest neighbors, and neural networks. These models are trained on labeled data and can handle numeric or categorical features, often requiring preprocessing such as normalization, encoding of categorical variables, and feature selection.

Performance is commonly measured with accuracy, along with class-wise metrics such as precision, recall, and F1 score. Confusion matrices and macro- or micro-averaged metrics help diagnose imbalances among classes. Cross-validation is frequently used to estimate generalization performance.

Applications include text classification (topic labeling, spam filtering), image or audio recognition with a single dominant category, and medical or industrial diagnostic tasks. A limitation is that the single-label assumption may be inappropriate when an instance truly belongs to multiple categories, or when label dependencies exist. In such cases, multi-label or structured prediction methods are used.