Trains a model possibly based on space-time folds

This function is basically a wrapper around functionality from the CAST and caret packages. It expects as input a data.frame object with the predictors and outcome variables present and well as possible variables indicating the spatial and temporal sub-groups of the training data. Either one of the spatial or temporal variables can be used to create individual space or time folds

train_model(
  traindata,
  predictors,
  response,
  spacevar = NULL,
  timevar = NULL,
  k = 10,
  ffs = TRUE,
  method = "rf",
  metric = ifelse(is.factor(response) | is.character(response), "Accuracy", "RMSE"),
  maximize = ifelse(metric == "RMSE", FALSE, TRUE),
  trControl = caret::trainControl(),
  seed = 42,
  verbose = TRUE,
  ...
)

Arguments

traindata	A tibble, data.frame or sf object.
predictors	A character vector with the column names used as predictors.
response	A character vector with the name of the response variable.
spacevar	An optional character with a variable indicating spatial groups to account for during training.
timevar	An optional character with a variable indicating temporal groups to account for during training.
k	An integer value indicating the number of folds. Defaults to 10.
ffs	A logical indicating if a Forward-Feature-Selection should be conducted. Note that this could significantly increase training time.
method	A charachter vector indicating a model to be used by caret. Defaults to "rf".
metric	An accuracy metric for model optimization. Defaults to Accuracy for classification and RMSE for regression. Must be supported by caret.
maximize	A logical indicating if the selected metric should be maximized. Defaults to TRUE for Accuracy and FALSE for RMSE.
trControl	A train control object based on trainControl.
seed	An integer value used to ensure reproducibility.
verbose	A logical indicating the level of verbosity.
...	Additional parameters used by ffs and train.

Value

A list with the model, performance metrics or confusion matrix (for regression or classification), a vector of observed values and a vector of predicted values.

Author

Darius Görgen (MapTailor Geospatial Consulting GbR) info@maptailor.net
Maintainer: MAPME-Initiative contact@mapme-initiative.org
Contact Person: Dr. Johannes Schielein
Copyright: MAPME-Initiative
License: GPL-3