1055linear124

1055linear124 is a synthetic benchmark dataset used in statistical learning and machine learning research to evaluate linear modeling techniques. The designation reflects its core dimensions and structure: 1055 samples and 124 features, with data generated under a linear data-generating process.

Construction: The feature matrix X is drawn from a standard normal distribution and standardized. The true coefficient vector beta is sparse, with a minority of features carrying nonzero weights, while the rest are zero. The response is formed as y = X beta plus Gaussian noise, with noise variance chosen to produce a moderate signal-to-noise ratio. This setup produces a controlled yet realistic setting for assessing regression performance and variable selection.

Usage and relevance: 1055linear124 serves as a common testbed for comparing estimation methods such as ordinary least squares, ridge regression, Lasso, and elastic net, as well as feature selection algorithms. It is also used to study the effects of noise, multicollinearity, and sparsity on predictive accuracy and model interpretability.

Variants and availability: Several variants of the dataset exist, differing in noise level, sparsity, or feature correlation structures. The dataset and accompanying reference implementations are typically hosted in public repositories with permissive licenses to facilitate reproducibility and benchmarking.