Orthogonal subsampling for big data linear regression

The dramatic growth of big datasets presents a new challenge to data storage and analysis. Data reduction, or subsampling, that extracts useful information from is crucial step in big-data We propose an orthogonal subsampling (OSS) approach for with focus on linear regression models. inspired by the fact array two levels provides best experimental design models sense it minimizes average variance estimated parameters predictions. merits OSS are three-fold: (i) easy implement fast; (ii) suitable distributed parallel computing ensures subsamples selected different batches have no common points, (iii) outperforms existing methods minimizing mean squared errors maximizing efficiencies subsamples. Theoretical results extensive numerical show superior approaches. It also more robust presence interactions among covariates, and, when they do exist, precise estimates interaction effects than methods. advantages illustrated through analysis real data.