![Error analysis and variable significance with random forests Ned Horning American Museum of Natural History's Center for Biodiversity and Conservation [email_address]](https://image.slidesharecdn.com/erroranalysisrandomforest-110426054854-phpapp01/85/Error-analysis-randomforest-1-320.jpg)
Error analysis randomforest
- 1. Error analysis and variable significance with random forests Ned Horning American Museum of Natural History's Center for Biodiversity and Conservation [email_address]
- 2. Error estimate Provides and unbiased estimate of the error Each tree uses a different bootstrap sample (~1/3 of samples) for testing Use function print() for OOB error estimate Use plot() to view plot of error estimate vs. number of trees Error rate vs. number of trees
- 3. Calculate OOB error estimate Put OOB samples down tree after it is constructed and keep track of results Proportion of times the result is not accurate averaged over all samples is the OOB error estimate For regression “percent variance explained” is also called pseudo R-squared OOB estimate of error rate: 0.1% Confusion matrix:
- 4. Variable importance Put oob samples down a tree then for each variable randomly reorder that variable in each of the oob samples and put these down the trees Two types of error can be calculated: mean decrease in accuracy and mean decrease in node impurity Actual measures depend if it is classification or regression
- 5. Variable importance steps In the randomForest() function specify “importance=TRUE” importance() function creates an importance object varImpPlot() function plots variable importance Specify type = 1 for mean decrease in accuracy and 2 for mean decrease in node impurity
- 6. Proximity measure Measures how frequent unique pairs of training samples (in and out of bag) end up in the same terminal node Used to fill in missing data and calculating outliers In the randomForest() function specify proximity=TRUE Outliers for classification
- 7. Outlier plots Use outlier() function to calculate outlier measures Can plot using the R plot() function Plot shows which samples contain variables that are outliers Outliers for classification