Use of spatialRF::auto_cor() and spatialRF::auto_vif() to reduce multicollinearity in a set of predictors while taking into account a preference order defined by the user..

multicollinearity.R

#installing and loading package
remotes::install_github(
  repo = "blasbenito/spatialRF", 
  ref = "main",
  force = TRUE,
  quiet = TRUE
)
library(spatialRF)
library(magrittr)

#loading other required pa
data(plant_richness_df)

#data frame with predictors
predictors.df <- plant_richness_df[, 5:21]

#1. manual preference order
#--------------------------------------

#preference order
preference.order <- c(
  "climate_bio1_average_X_bias_area_km2",
  "climate_aridity_index_average",
  "climate_hypervolume",
  "climate_bio1_average",
  "climate_bio15_minimum",
  "bias_area_km2"
)

#reducing multicollinearity
#via Pearson correlation with auto_cor()
#and variance inflation factor via auto_vif()
variable.selection <- spatialRF::auto_cor(
  x = predictors.df,
  cor.threshold = 0.5,
  preference.order = preference.order
) %>% 
  spatialRF::auto_vif(
    vif.threshold = 2.5,
    preference.order = preference.order
  )

#selected variables
variable.selection$selected.variables



#applying auto_vif() before auto_cor()
#does not change the result
variable.selection <- spatialRF::auto_vif(
  x = predictors.df,
  vif.threshold = 2.5,
  preference.order = preference.order
) %>% 
  spatialRF::auto_cor(
  cor.threshold = 0.5,
  preference.order = preference.order
) 

#selected variables
variable.selection$selected.variables


#2. automatic preference order
#------------------------------------

#fitting a model to extract importance scores
m <- spatialRF::rf(
  data = plant_richness_df,
  dependent.variable.name = "richness_species_vascular",
  predictor.variable.names = colnames(predictors.df),
  verbose = FALSE
)

#plotting importance scores
spatialRF::plot_importance(m)

#getting preference order in order of variable importance
preference.order <- m$importance$per.variable$variable

#reducing multicollinearity
variable.selection <- spatialRF::auto_cor(
  x = predictors.df,
  cor.threshold = 0.5,
  preference.order = preference.order
) %>% 
  spatialRF::auto_vif(
    vif.threshold = 2.5,
    preference.order = preference.order
  )

#selected variables
variable.selection$selected.variables