Get Your Data Ready For Machine Learning in R with Pre
Preparing data is required to get the best results from machine learning algorithms.
In this post you will discover how to transform your data in order to best expose its structure to machine learning algorithms in R using the caret package.
You will work through 8 popular and powerful data transforms with recipes that you can study or copy and paste int your current or next machine learning project.
Let’s get started.
Need For Data Pre-Processing
You want to get the best accuracy from machine learning algorithms on your datasets.
Some machine learning algorithms require the data to be in a specific form. Whereas other algorithms can perform better if the data is prepared in a specific way, but not always. Finally, your raw data may not be in the best format to best expose the underlying structure and relationships to the predicted variables.
It is important to prepare your data in such a way that it gives various different machine learning algorithms the best chance on your problem.
You need to pre-process your raw data as part of your machine learning project.
Data Pre-Processing Methods
It is hard to know which data-preprocessing methods to use.
You can use rules of thumb such as:
- Instance based methods are more effective if the input attributes have the same scale.
- Regression methods can work better of the input attributes are standardized.
These are heuristics, but not hard and fast laws of machine learning, because sometimes you can get better results if you ignore them.
You should try a range of data transforms with a range of different machine learning algorithms. This will help you discover both good representations for your data and algorithms that are better at exploiting the structure that those representations expose.
It is a good idea to spot check a number of transforms both in isolation as well as combinations of transforms.
In the next section you will discover how you can apply data transforms in order to prepare your data in R using the caret package.
Need more Help with R for Machine Learning?
Take my free 14-day email course and discover how to use R on your project (with sample code).
Click to sign-up and also get a free PDF Ebook version of the course.
Data Pre-Processing With Caret in R
The caret package in R provides a number of useful data transforms.
These transforms can be used in two ways.
- Standalone: Transforms can be modeled from training data and applied to multiple datasets. The model of the transform is prepared using the preProcess() function and applied to a dataset using the predict() function.
- Training: Transforms can prepared and applied automatically during model evaluation. Transforms applied during training are prepared using the preProcess() and passed to the train() function via the preProcess argument.
A number of data preprocessing examples are presented in this section. They are presented using the standalone method, but you can just as easily use the prepared preprocessed model during model training.
All of the preprocessing examples in this section are for numerical data. Note that the preprocessing functions will skip over non-numeric data without raising an error.
You can learn more about the data transforms provided by the caret package by reading the help for the preProcess function by typing ?preProcess and by reading the Caret Pre-Processing page.
The data transforms presented are more likely to be useful for algorithms such as regression algorithms, instance-based methods (like kNN and LVQ), support vector machines and neural networks. They are less likely to be useful for tree and rule based methods.
Summary of Transform Methods
Below is a quick summary of all of the transform methods supported in the method argument of the preProcess() function in caret.
- “BoxCox“: apply a Box–Cox transform, values must be non-zero and positive.
- “YeoJohnson“: apply a Yeo-Johnson transform, like a BoxCox, but values can be negative.
- “expoTrans“: apply a power transform like BoxCox and YeoJohnson.
- “zv“: remove attributes with a zero variance (all the same value).
- “nzv“: remove attributes with a near zero variance (close to the same value).
- “center“: subtract mean from values.
- “scale“: divide values by standard deviation.
- “range“: normalize values.
- “pca“: transform data to the principal components.
- “ica“: transform data to the independent components.
- “spatialSign“: project data onto a unit circle.
The following sections will demonstrate some of the more popular methods.
1. Scale
The scale transform calculates the standard deviation for an attribute and divides each value by that standard deviation.
1234567891011121314 | # load librarieslibrary(caret)# load the datasetdata(iris)# summarize datasummary(iris[,1:4])# calculate the pre-process parameters from the datasetpreprocessParams<-preProcess(iris[,1:4],method=c("scale"))# summarize transform parametersprint(preprocessParams)# transform the dataset using the parameterstransformed<-predict(preprocessParams,iris[,1:4])# summarize the transformed datasetsummary(transformed) |
Running the recipe, you will see:
123456789101112131415161718192021 | Sepal.Length Sepal.Width Petal.Length Petal.Width Min. :4.300 Min. :2.000 Min. :1.000 Min. :0.100 1st Qu.:5.100 1st Qu.:2.800 1st Qu.:1.600 1st Qu.:0.300 Median :5.800 Median :3.000 Median :4.350 Median :1.300 Mean :5.843 Mean :3.057 Mean :3.758 Mean :1.199 3rd Qu.:6.400 3rd Qu.:3.300 3rd Qu.:5.100 3rd Qu.:1.800 Max. :7.900 Max. :4.400 Max. :6.900 Max. :2.500 Created from 150 samples and 4 variablesPre-processing: - ignored (0) - scaled (4) Sepal.Length Sepal.Width Petal.Length Petal.Width Min. :5.193 Min. : 4.589 Min. :0.5665 Min. :0.1312 1st Qu.:6.159 1st Qu.: 6.424 1st Qu.:0.9064 1st Qu.:0.3936 Median :7.004 Median : 6.883 Median :2.4642 Median :1.7055 Mean :7.057 Mean : 7.014 Mean :2.1288 Mean :1.5734 3rd Qu.:7.729 3rd Qu.: 7.571 3rd Qu.:2.8890 3rd Qu.:2.3615 Max. :9.540 Max. :10.095 Max. :3.9087 Max. :3.2798 |
2. Center
The center transform calculates the mean for an attribute and subtracts it from each value.
1234567891011121314 | # load librarieslibrary(caret)# load the datasetdata(iris)# summarize datasummary(iris[,1:4])# calculate the pre-process parameters from the datasetpreprocessParams<-preProcess(iris[,1:4],method=c("center"))# summarize transform parametersprint(preprocessParams)# transform the dataset using the parameterstransformed<-predict(preprocessParams,iris[,1:4])# summarize the transformed datasetsummary(transformed) |
Running the recipe, you will see:
123456789101112131415161718192021 | Sepal.Length Sepal.Width Petal.Length Petal.Width Min. :4.300 Min. :2.000 Min. :1.000 Min. :0.100 1st Qu.:5.100 1st Qu.:2.800 1st Qu.:1.600 1st Qu.:0.300 Median :5.800 Median :3.000 Median :4.350 Median :1.300 Mean :5.843 Mean :3.057 Mean :3.758 Mean :1.199 3rd Qu.:6.400 3rd Qu.:3.300 3rd Qu.:5.100 3rd Qu.:1.800 Max. :7.900 Max. :4.400 Max. :6.900 Max. :2.500 Created from 150 samples and 4 variablesPre-processing: - centered (4) - ignored (0) Sepal.Length Sepal.Width Petal.Length Petal.Width Min. :-1.54333 Min. :-1.05733 Min. :-2.758 Min. :-1.0993 1st Qu.:-0.74333 1st Qu.:-0.25733 1st Qu.:-2.158 1st Qu.:-0.8993 Median :-0.04333 Median :-0.05733 Median : 0.592 Median : 0.1007 Mean : 0.00000 Mean : 0.00000 Mean : 0.000 Mean : 0.0000 3rd Qu.: 0.55667 3rd Qu.: 0.24267 3rd Qu.: 1.342 3rd Qu.: 0.6007 Max. : 2.05667 Max. : 1.34267 Max. : 3.142 Max. : 1.3007 |
3. Standardize
Combining the scale and center transforms will standardize your data. Attributes will have a mean value of 0 and a standard deviation of 1.
1234567891011121314 | # load librarieslibrary(caret)# load the datasetdata(iris)# summarize datasummary(iris[,1:4])# calculate the pre-process parameters from the datasetpreprocessParams<-preProcess(iris[,1:4],method=c("center","scale"))# summarize transform parametersprint(preprocessParams)# transform the dataset using the parameterstransformed<-predict(preprocessParams,iris[,1:4])# summarize the transformed datasetsummary(transformed) |
Notice how we can list multiple methods in a list when defining the preProcess procedure in caret. Running the recipe, you will see:
12345678910111213141516171819202122 | Sepal.Length Sepal.Width Petal.Length Petal.Width Min. :4.300 Min. :2.000 Min. :1.000 Min. :0.100 1st Qu.:5.100 1st Qu.:2.800 1st Qu.:1.600 1st Qu.:0.300 Median :5.800 Median :3.000 Median :4.350 Median :1.300 Mean :5.843 Mean :3.057 Mean :3.758 Mean :1.199 3rd Qu.:6.400 3rd Qu.:3.300 3rd Qu.:5.100 3rd Qu.:1.800 Max. :7.900 Max. :4.400 Max. :6.900 Max. :2.500 Created from 150 samples and 4 variablesPre-processing: - centered (4) - ignored (0) - scaled (4) Sepal.Length Sepal.Width Petal.Length Petal.Width Min. :-1.86378 Min. :-2.4258 Min. :-1.5623 Min. :-1.4422 1st Qu.:-0.89767 1st Qu.:-0.5904 1st Qu.:-1.2225 1st Qu.:-1.1799 Median :-0.05233 Median :-0.1315 Median : 0.3354 Median : 0.1321 Mean : 0.00000 Mean : 0.0000 Mean : 0.0000 Mean : 0.0000 3rd Qu.: 0.67225 3rd Qu.: 0.5567 3rd Qu.: 0.7602 3rd Qu.: 0.7880 Max. : 2.48370 Max. : 3.0805 Max. : 1.7799 Max. : 1.7064 |
4. Normalize
Data values can be scaled into the range of [0, 1] which is called normalization.
1234567891011121314 | # load librarieslibrary(caret)# load the datasetdata(iris)# summarize datasummary(iris[,1:4])# calculate the pre-process parameters from the datasetpreprocessParams<-preProcess(iris[,1:4],method=c("range"))# summarize transform parametersprint(preprocessParams)# transform the dataset using the parameterstransformed<-predict(preprocessParams,iris[,1:4])# summarize the transformed datasetsummary(transformed) |
Running the recipe, you will see:
12345678910111213141516171819202122 | Sepal.Length Sepal.Width Petal.Length Petal.Width Min. :4.300 Min. :2.000 Min. :1.000 Min. :0.100 1st Qu.:5.100 1st Qu.:2.800 1st Qu.:1.600 1st Qu.:0.300 Median :5.800 Median :3.000 Median :4.350 Median :1.300 Mean :5.843 Mean :3.057 Mean :3.758 Mean :1.199 3rd Qu.:6.400 3rd Qu.:3.300 3rd Qu.:5.100 3rd Qu.:1.800 Max. :7.900 Max. :4.400 Max. :6.900 Max. :2.500 Created from 150 samples and 4 variablesPre-processing: - ignored (0) - re-scaling to [0, 1] (4) Sepal.Length Sepal.Width Petal.Length Petal.Width Min. :0.0000 Min. :0.0000 Min. :0.0000 Min. :0.00000 1st Qu.:0.2222 1st Qu.:0.3333 1st Qu.:0.1017 1st Qu.:0.08333 Median :0.4167 Median :0.4167 Median :0.5678 Median :0.50000 Mean :0.4287 Mean :0.4406 Mean :0.4675 Mean :0.45806 3rd Qu.:0.5833 3rd Qu.:0.5417 3rd Qu.:0.6949 3rd Qu.:0.70833 Max. :1.0000 Max. :1.0000 Max. :1.0000 Max. :1.00000 |
5. Box-Cox Transform
When an attribute has a Gaussian-like distribution but is shifted, this is called a skew. The distribution of an attribute can be shifted to reduce the skew and make it more Gaussian. The BoxCox transform can perform this operation (assumes all values are positive).
123456789101112131415 | # load librarieslibrary(mlbench)library(caret)# load the datasetdata(PimaIndiansDiabetes)# summarize pedigree and agesummary(PimaIndiansDiabetes[,7:8])# calculate the pre-process parameters from the datasetpreprocessParams< |