Prism: Multiple spline regression with regularization, dimensionality reduction, and feature selection
Christopher R. Madan · The Journal of Open Source Software · 2016
Prism uses a combination of statistical methods to conduct spline-based multiple regression.Prism conducts this regression using regularization, dimensionality reduction, and feature selection, through a combination of smoothing spline regression, PCA, and RVR/LASSO.Smoothing splines can be used to model non-parametric relationships using piece-wise cubic functions (Wahba and Wold 1975;Fox 2000).Relevance vector regression (RVR) refers to application of a relevance vector machine (RVM) to a regression problem; broadly, RVM is similar to multiple linear regression with regularization, using automatic relevance determination for feature selection (Tipping 2000).RVM shares many commonalities with SVM, and is implemented as a special case of a Sparse Bayesian framework (Tipping 2001;Tipping and Faul 2003).