Bot Botany – K-Means and ggplot2

[This article was first published on R-Chart, 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.

So if you had a robot that was an expert at botany – would you have a bot botanist?  Among other things, it would need to to distinguish flowers through vision and image processing, and be able to classify various kinds of plants based upon specific characteristics.  What do both of these requirements have in common?   They can be done using the k-means clustering.  Image segmentation can be used to allow our robot to recognize objects.  Based upon petal and sepal size, it could determine say – the species of an iris.  The well-known iris data set has been featured in other posts.

K-Means in R
If you look up the k-means algorithm online or in a reference book, you will be met with a flurry a mathematical symbols and formal explanations.  The basic principal (informally stated) is rather simple… given set of observations (picture a scatter plot of points), and a number of groups or clusters that you wish to group them in, the k-means algorithm finds the center of each group and associates observations with the groups with the “closest” center.

To use k-means in R, call the kmeans function with a matrix of values and the number of centers.  The function seeks to partition the points into k groups (the number of centers) such that the sum of squares from points to the assigned cluster centers is minimized.  Each observation (point) belongs to the cluster with the nearest mean.

To start, we will copy the iris data set to a separate data frame.  Not strictly speaking necessary, but makes it easier me to reflexively enter df whenever the data frame is in view.  Next we create a matrix object containing only the Petal Length and Width.

m=as.matrix(cbind(df$Petal.Length, df$Petal.Width),ncol=2)

Now we will do the actual clustering.


Simple eh?  The cl object contains a number of interesting attributes associated with the model.


Next we do a bit of data formatting and preparation for subsequent calls to graph the data.  Notice that we add the cluster information back to our original data frame.  This is a good organization of the data and also a requirement for working with ggplot2 which is designed to use data frames.


The following graph color codes the points by cluster.  We also add the centers and a semi transparent halo around the center to emphasize the place of the center… and its role in classifying the observations into clusters.


ggplot(data=df, aes(x=Petal.Length, y=Petal.Width, color=cluster )) + 
 geom_point() + 
 geom_point(data=centers, aes(x=V1,y=V2, color=’Center’)) +
 geom_point(data=centers, aes(x=V1,y=V2, color=’Center’), size=52, alpha=.3, legend=FALSE)

This plot is an interesting example of how several different sets of data (in this case the actual observations as well as the centers) in separate data frames can be included in a single ggplot2 chart.

Misclassified Observations
The models is pretty accurate, but not perfect.  The following SQL statement highlights the few misclassified observations:

sqldf(‘select Species, cluster, count(*) from df group by Species, Cluster’)

     Species cluster count(*)
1     setosa       2       50
2 versicolor       1       48
3 versicolor       3        2
4  virginica       1        6
5  virginica       3       44

So we grab the outliers into their own data frame….

df2 = sqldf(‘select * from df where (Species || cluster) in 
             (select Species || cluster from df group by Species, Cluster having count(*) < 10)')

Now we can enhance the previous graph to put a diamond around misclassified points.

last_plot() +  geom_point(data=df2, aes(x=Petal_Length, y=Petal_Width, shape=5, alpha=.7, size=4.5), legend=FALSE) 

And so with a bit of Data Mining knowledge and the R programming language, even our machines can stop and smell the roses…

To leave a comment for the author, please follow the link and comment on their blog: R-Chart. 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)