fRFE

Function to [R]ecursive [F]eature [E]limination for the prediction of soil classes and numeric parameters

Source

https://github.com/FLFgit/ScaleP/blob/master/functions/fRFE.R

Parameter

• RU.DIR -> directory containing reference unit (RU) file
• RU.SHP -> name of reference unit shape file
• SAMPLE.DIR -> directory containing sample data set
• SAMPLE.SHP -> name of sample data set
• PART -> proportion of training and test data set [0...1]
• OUT.DIR -> directory containing resulting files
• EPSG -> EPSG code of all data
• T.PM -> column name of target parameter in SAMPLE.SHP
• PM -> prefix of explaning attributes, which should be considered
• CLASS=TRUE -> option: classifiation is TRUE
• UPTRAIN=TRUE -> option: if CLASS=TRUE, UPTRAIN function for balancing of unequally distributed classes is TRUE
• NORMALIZATION=TRUE -> option: normalization of explaining attributes

Result

• [RU.SHP]-[SAMPLE.SHP]_[T.PM]_RFE-BV.txt -> list of best performing variables
• [RU.SHP]-[SAMPLE.SHP]_[T.PM]_RFE-OA.txt -> model overall accuracy
• [RU.SHP]-[SAMPLE.SHP]_[T.PM]_RFE-IMP.csv -> variable importance of best performing variables
• [RU.SHP]-[SAMPLE.SHP]_[T.PM]_RFE-ACC.csv -> variable-specific accuracy metrics
• [RU.SHP]-[SAMPLE.SHP]_[T.PM]_RFE-ACC.pdf -> accuracy metric plot depending on variable combinations

Numeric prediction

• [RU.SHP]-[SAMPLE.SHP]_[T.PM]_RFE-DP.pdf -> density end ECDF plots of training and test data sets

Classification

• [RU.SHP]-[SAMPLE.SHP]_[T.PM]_RFE-BP.pdf -> barplot of class proportions for test and training data
• [RU.SHP]-[SAMPLE.SHP]_[T.PM]_RFE-CV.csv -> confusion matrix