Datensatz

Datensatz, in English dataset, is a collection of data records used for analysis. In statistics and data science, a Datensatz consists of rows (observations) and columns (variables or attributes). Each row represents one instance; each column a feature. Datasets may be structured as tables in CSV or SQL databases, semi-structured as JSON or XML, or, less commonly, unstructured but organized for processing. Metadata describes the dataset, including its name, source, collection date, units, and data quality.

In machine learning, a dataset often includes features and a target label. Datasets are partitioned into training, validation, and test sets. They can be small or large, static or streaming, and may be stored in various formats or systems such as files, databases, or data lakes.

Quality and preparation are important: collection, cleaning, normalization, deduplication, and annotation. Provenance information records how the data were obtained and transformed. Reproducibility benefits from clear documentation, versioning, and stable storage.

Ethics and law: privacy, anonymization, consent, and licensing govern sharing and reuse. Open datasets may be released under licenses like CC0 or CC BY, while sensitive data require access controls.

Related concepts include data frames, matrices, and data collections. A well-documented Datensatz supports reliable analysis, transparent methods, and reproducible research.