Principal component analysis (PCA) in R

[This article was first published on Methods – finnstats, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

PCA is used in exploratory data analysis and for making decisions in predictive models.

PCA commonly used for dimensionality reduction by using each data point onto only the first few principal components (most cases first and second dimensions) to obtain lower-dimensional data while keeping as much of the data’s variation as possible.

The first principal component can equivalently be defined as a direction that maximizes the variance of the projected data.

The principal components are often analyzed by eigendecomposition of the data covariance matrix or singular value decomposition (SVD) of the data matrix.

Decision Trees in R

Reducing the number of variables from a data set naturally leads to inaccuracy, but the trick in the dimensionality reduction is to allow us to make correct decisions based on high accuracy.

Always smaller data sets are easier to explore, visualize, analyze, and faster for machine learning algorithms.

In this tutorial we will make use of iris dataset in R for analysis & interprettion.

Self Organizing Maps

Getting Data


data.frame’:     150 obs. of  5 variables:

 $ Sepal.Length: num  5.1 4.9 4.7 4.6 5 5.4 4.6 5 4.4 4.9 ...
 $ Sepal.Width : num  3.5 3 3.2 3.1 3.6 3.9 3.4 3.4 2.9 3.1 ...
 $ Petal.Length: num  1.4 1.4 1.3 1.5 1.4 1.7 1.4 1.5 1.4 1.5 ...
 $ Petal.Width : num  0.2 0.2 0.2 0.2 0.2 0.4 0.3 0.2 0.2 0.1 ...
 $ Species     : Factor w/ 3 levels "setosa","versicolor",..: 1 1 1 1 1 1 1 1 1 1 ...

In this datasets contains 150 observations with 5 variables.

Sepal.Length  Sepal.Width   Petal.Length  Petal.Width        Species 
 Min.   :4.3   Min.   :2.0   Min.   :1.0   Min.   :0.1   setosa    :50 
 1st Qu.:5.1   1st Qu.:2.8   1st Qu.:1.6   1st Qu.:0.3   versicolor:50 
 Median :5.8   Median :3.0   Median :4.3   Median :1.3   virginica :50 
 Mean   :5.8   Mean   :3.1   Mean   :3.8   Mean   :1.2                 
 3rd Qu.:6.4   3rd Qu.:3.3   3rd Qu.:5.1   3rd Qu.:1.8                 
 Max.   :7.9   Max.   :4.4   Max.   :6.9   Max.   :2.5 

Partition Data

Lets divides the data sets into training dataset and test datasets.

Exploratory Data Analysis in R

ind <- sample(2, nrow(iris),
              replace = TRUE,
              prob = c(0.8, 0.2))
training <- iris[ind==1,]
testing <- iris[ind==2,]

Scatter Plot & Correlations


First will check the correlation between independent variables. Let’s remove the factor variable from the dataset for correlation data analysis.

KNN Machine Algorithm in R

             gap = 0,
             bg = c("red", "yellow", "blue")[training$Species],

Lower triangles provide scatter plots and upper triangles provide correlation values.

Petal length and petal width are highly correlated. Same way sepal length and petal length , Sepeal length, and petal width are also highly correlated.

This leads to multicollinearity issues. So if we predict the model based on this dataset may be erroneous.

One way handling these kinds of issues is based on PCA.

Cluster optimization in R

Principal Component Analysis

Principal Component Analysis is based on only independent variables. So we removed the fifth variable from the dataset.

pc <- prcomp(training[,-5],
             center = TRUE,
            scale. = TRUE)


[1] "sdev"     "rotation" "center" 
[4] "scale"    "x"      
[1] "prcomp"
Sepal.Length  Sepal.Width Petal.Length
         5.8          3.1          3.6

Scale function is used for normalization

Sepal.Length  Sepal.Width Petal.Length
        0.82         0.46         1.79

Print the results stored in pc.


while printing pc you will get standard deviations and loadings.

Standard deviations (1, .., p=4):
[1] 1.72 0.94 0.38 0.14
Rotation (n x k) = (4 x 4):
               PC1    PC2   PC3   PC4
Sepal.Length  0.51 -0.398  0.72  0.23
Sepal.Width  -0.29 -0.913 -0.26 -0.12
Petal.Length  0.58 -0.029 -0.18 -0.80
Petal.Width   0.56 -0.081 -0.62  0.55

For example, PC1 increases when Sepal Length, Petal Length, and Petal Width are increased and it is positively correlated whereas PC1 increases Sepal Width decrease because these values are negatively correlated.

Naïve Bayes Classification in R

Importance of components:
                         PC1   PC2    PC3
Standard deviation     1.717 0.940 0.3843
Proportion of Variance 0.737 0.221 0.0369
Cumulative Proportion  0.737 0.958 0.9953
Standard deviation     0.1371
Proportion of Variance 0.0047
Cumulative Proportion  1.0000

The first principal components explain the variability around 73% and its captures the majority of the variability.

In this case, the first two components capture the majority of the variability.

Orthogonality of PCs

Let’s create the scatterplot based on PC and see the multicollinearity issue is addressed or not?.

             bg = c("red", "yellow", "blue")[training$Species],

Now the correlation coefficients are zero, so we can get rid of multicollinearity issues.


For making biplot need devtools package.

Share point and R Integration

g <- ggbiplot(pc,
              obs.scale = 1,
              var.scale = 1,
              groups = training$Species,
              ellipse = TRUE,
              circle = TRUE,
              ellipse.prob = 0.68)
g <- g + scale_color_discrete(name = '')
g <- g + theme(legend.direction = 'horizontal',
               legend.position = 'top')

PC1 explains about 73.7% and PC2 explained about 22.1% of variability.

Arrows are closer to each other indicates the high correlation.

For example correlation between Sepal Width vs other variables is weakly correlated.

Another way interpreting the plot is PC1 is positively correlated with the variables Petal Length, Petal Width, and Sepal Length, and PC1 is negatively correlated with Sepal Width.

PC2 is highly negatively correlated with Sepal Width.

Bi plot is an important tool in PCA to understand what is going on in the dataset.

Difference between association and correlation

Prediction with Principal Components

trg <- predict(pc, training)

Add the species column information into trg.

trg <- data.frame(trg, training[5])
tst <- predict(pc, testing)
tst <- data.frame(tst, testing[5])

Multinomial Logistic regression with First Two PCs

Because our dependent variable has 3 level, so we will make use of multinomial logistic regression.

trg$Species <- relevel(trg$Species, ref = "setosa")
mymodel <- multinom(Species~PC1+PC2, data = trg)

Model out is given below and we used only first two principal components, because majority of the information’s available in first components.

How to measure quality control of the product?

multinom(formula = Species ~ PC1 + PC2, data = trg)
           (Intercept) PC1 PC2
versicolor        7.23  14 3.2
virginica        -0.58  20 3.6
Std. Errors:
           (Intercept) PC1 PC2
versicolor         188 106 128
virginica          188 106 128
Residual Deviance: 36
AIC: 48

Confusion Matrix & Misclassification Error – training

p <- predict(mymodel, trg)
tab <- table(p, trg$Species)

Lets see the correct classification and miss classifications in training dataset.

Stock Prediction in R

   p          setosa versicolor virginica
  setosa         45          0         0
  versicolor      0         35         3
  virginica       0          5        32
1 - sum(diag(tab))/sum(tab)

Misclassification error is 0.067

Confusion Matrix & Misclassification Error – testing

p1 <- predict(mymodel, tst)
tab1 <- table(p1, tst$Species)

Based on the test data set error classification is calculated.

p1           setosa versicolor virginica
  setosa          5          0         0
  versicolor      0          9         3
  virginica       0          1        12
1 - sum(diag(tab1))/sum(tab1)

Misclassification for the testing data set is 13.33%.

Class Imbalance Handling in R

The post Principal component analysis (PCA) in R appeared first on finnstats.

To leave a comment for the author, please follow the link and comment on their blog: Methods – finnstats. offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)