fof_pc {funcharts}

Function-on-function linear regression based on principal components

Description

Function-on-function linear regression based on principal components. This function performs multivariate functional principal component analysis (MFPCA) to extract multivariate functional principal components from the multivariate functional covariates as well as from the functional response, then it builds a linear regression model of the response scores on the covariate scores. Both functional covariates and response are standardized before the regression. See Centofanti et al. (2021) for additional details.

Usage

fof_pc(
  mfdobj_y,
  mfdobj_x,
  tot_variance_explained_x = 0.95,
  tot_variance_explained_y = 0.95,
  tot_variance_explained_res = 0.95,
  components_x = NULL,
  components_y = NULL,
  type_residuals = "standard"
)

Arguments

mfdobj_y

A multivariate functional data object of class mfd denoting the functional response variable. Although it is a multivariate functional data object, it must have only one functional variable.

mfdobj_x

A multivariate functional data object of class mfd denoting the functional covariates.

tot_variance_explained_x

The minimum fraction of variance that has to be explained by the multivariate functional principal components retained into the MFPCA model fitted on the functional covariates. Default is 0.95.

tot_variance_explained_y

The minimum fraction of variance that has to be explained by the multivariate functional principal components retained into the MFPCA model fitted on the functional response. Default is 0.95.

tot_variance_explained_res

The minimum fraction of variance that has to be explained by the multivariate functional principal components retained into the MFPCA model fitted on the functional residuals of the functional regression model. Default is 0.95.

components_x

A vector of integers with the components over which to project the functional covariates. If NULL, the first components that explain a minimum fraction of variance equal to tot_variance_explained_x is selected. #' If this is not NULL, the criteria to select components are ignored. Default is NULL.

components_y

A vector of integers with the components over which to project the functional response. If NULL, the first components that explain a minimum fraction of variance equal to tot_variance_explained_y is selected. #' If this is not NULL, the criteria to select components are ignored. Default is NULL.

type_residuals

A character value that can be "standard" or "studentized". If "standard", the MFPCA on functional residuals is calculated on the standardized covariates and response. If "studentized", the MFPCA on studentized version of the functional residuals is calculated on the non-standardized covariates and response. See Centofanti et al. (2021) for additional details.

Value

A list containing the following arguments:

• mod: an object of class lm that is a linear regression model where the response variables are the MFPCA scores of the response variable and the covariates are the MFPCA scores of the functional covariates. mod$coefficients contains the matrix of coefficients of the functional regression basis functions,

• beta_fd: a bifd object containing the bivariate functional regression coefficients \beta(s,t) estimated with the function-on-function linear regression model,

• fitted.values: a multivariate functional data object of class mfd with the fitted values of the functional response observations based on the function-on-function linear regression model,

• residuals_original_scale: a multivariate functional data object of class mfd with the functional residuals of the function-on-function linear regression model on the original scale, i.e. they are the difference between mfdobj_y and fitted.values,

• residuals: a multivariate functional data object of class mfd with the functional residuals of the function-on-function linear regression model, standardized or studentized depending on the argument type_residuals,

• type_residuals: the same as the provided argument,

• pca_x: an object of class pca_mfd obtained by doing MFPCA on the functional covariates,

• pca_y: an object of class pca_mfd obtained by doing MFPCA on the functional response,

• pca_res: an object of class pca_mfd obtained by doing MFPCA on the functional residuals,

• components_x: a vector of integers with the components selected in the pca_x model,

• components_y: a vector of integers with the components selected in the pca_y model,

• components_res: a vector of integers with the components selected in the pca_res model,

• y_standardized: the standardized functional response obtained doing scale_mfd(mfdobj_y),

• tot_variance_explained_x: the same as the provided argument

• tot_variance_explained_y: the same as the provided argument

• tot_variance_explained_res: the same as the provided argument

• get_studentized_residuals: a function that allows to calculate studentized residuals on new data, given the estimated function-on-function linear regression model.

References

Centofanti F, Lepore A, Menafoglio A, Palumbo B, Vantini S. (2021) Functional Regression Control Chart. Technometrics, 63(3), 281–294. doi:10.1080/00401706.2020.1753581

Examples

library(funcharts)
data("air")
air <- lapply(air, function(x) x[1:10, , drop = FALSE])
fun_covariates <- c("CO", "temperature")
mfdobj <- get_mfd_list(air, lambda = 1e-2)
mfdobj_y <- mfdobj[, "NO2"]
mfdobj_x <- mfdobj[, fun_covariates]
mod <- fof_pc(mfdobj_y, mfdobj_x)