Penalized Regression in Large-Scale Data Analysis

Among machine learning methods, penalized regression provides interpretable predictive models, which increases its importance and usability in educational research in which explanation has been valued over prediction. Particularly coupled with large-scale data, the sparsity assumption of penalized regression is most likely to be met, and penalized regression contributes to the exploration and identification of yet uninvestigated variables or relationships. Penalized regression methods such as LASSO, elastic net, and group Mnet have been widely applied to educational large-scale data. Recently, penalized regression has been extended to various statistical models, including significance testing and multilevel models. In this chapter, we will overview predictive modeling, and explain the basics of penalized regression and model assessment, followed by extensions of penalized regression. Finally, future research topics are addressed, and R coding examples are provided for reference in the context of educational large-scale data analysis.