Boosting for Estimating Spatially Structured Additive Models

Inference and computation with generalized additive models and their extensions

“…A rather different approach to smooth model estimation uses boosting (e.g. Tutz and Binder 2006;Schmid and Hothorn 2008;Robinzonov and Hothorn 2010;Mayr et al 2012). Boosting is a forward selection strategy, in which smooth model components are iteratively built up from over-smoothed versions, fitted to generalized residuals (Hastie and Tibshirani 1986) for additive model fitting.…”

EM and component-wise boosting for Hidden Markov Models: a machine-learning approach to capture-recapture

“…Statistical boosting can be thought of in two ways. One, it is an iterative method for obtaining a statistical model, G ( X ), via functional gradient descent (Breiman, 1998; Friedman et al, 2000; Friedman, 2001; Breiman, 1999; Schmid et al, 2010; Nikolay Robinzonov, 2013), where and is the fit-vector , the expected values of Y based on covariate data X . Although boosting has origins in classification algorithms, we now know that it is equivalent to regularized regression, such as the Lasso (Bühlmann & Yu, 2003; Efron et al, 2004, under certain conditions).…”

Extending Statistical Boosting

“…An extension to two-dimensional P-splines to include spatial effects or interactions was proposed by Kneib et al [32]. For an application incorporating also spatio-temporal effects see Robinzonov and Hothorn [33]. To include discrete spatial effects such as a regional structure, Sobotka and Kneib [34] proposed a Markov random field base-learner applying a penalization which ensures that neighboring regions share similar effects.…”