BestPredFFS: Select best performance layers for classification

In SchoenbergA/IKARUS: Classification Tool

Description

uses a forward feature selection (FFS) to select the best predictors for the classification

Usage

BestPredFFS(tDat, predCol = "default", classCol = NULL, Cores = 1)

Arguments

tDat	data.frame - with values of the predictors (see details)
predCol	numeric - seq of columns with the predictor values. By default uses 1:(length(tDat)-1) for tDat format computed by 'IKARUS::exrct_Traindat'
classCol	character - name of the column containing the class information
Cores	numeric - amount of Cores to exclude from calculation, default = 1

Details

The function is used to select best performing predictor variables for a classification. The

• predCol - specific predictors can be selected by setting predCol = x:y

• tDat - the use of IKARUS::exrct_Traindat is recommended.

• parallel processing - the function uses parallel processing for multicore processors. by default all cores -1 are used.

Value

returns a list of best performing predictors

Note

The function will compute a huge number of models. Depending on the sizes of the training data the process can take long time even with multicore processing.

Author(s)

Andreas Schönberg

Examples

# load data
require(caret)
require(CAST)
require(doParallel)
require(raster)
require(IKARUS)
lau_Stk <- raster::stack(system.file("extdata","lau_RGB.grd",package = "IKARUS"))
lau_tP <-rgdal::readOGR(system.file("extdata","lau_TrainPolygon.shp",package = "IKARUS"))
crs(lau_tP) <- crs(lau_Stk)
### extract values using 'IKARUS::exrct_Traindat'
tDat <- exrct_Traindat(lau_tP,lau_Stk,"class")
# FFS with all layers in the RasterStack (this example could take some minutes)
ffs <- BestPredFFS(tDat=tDat,classCol = "class")
# some code to look at the results
ffs$selectedvars # show seleted variables
ffs$perf_all # show performance of all combinations
ffs$finalModel # show confusion matrix

SchoenbergA/IKARUS documentation built on Sept. 8, 2021, 11:11 a.m.