How I build a classification model with R

Exploring and analysing of the Iris Dataset

Classification is a very important area of machine learning, as it allows you to create categories based on certain characteristics. It is used in a lot of fields nowadays such as marketing, where we can classify visitors of a sales site according to their appetite to buy. We were just talking about it in a previous article .

In this one we will take a set of flowers and their characteristics. We are going to analyze the different characteristics of these flowers and try to see if they are correlated. This will help us to create a classification model that each time we give the characteristics of a flower we can tell which species it is.

For this analysis we will use a dataset that comes from Kaggle a very famous dataset bank. This one indexes 150 flowers to which one associates a species and some characteristics.

Features Presentation

Let’s see what we can find in this dataset just by using the summary function.

We can see here that we have three different species: Setosa, Versicolor and Virginica with 50 records for each species. We can also see that petal length feature has a very high range compared to the others. If we go deeper, we can say that the high difference between the first quartile and the median could be caused by a species that has some very little petals. Enough speculation! Let’s dive into this dataset together.

Finding a discriminatory feature

It will be interesting to study which characteristic(s) discriminates each species and to what extent.

To do this we will start by analyzing each species in relation to the others, in order to determine if one of the characteristics is not in the average. If so, we will analyse this characteristic more precisely.

Setosa

We’ll begin by analysing if one of the features of the Setosa is far from the mean.

We begin by saving the mean of the features concerning the Setosa species in one variable and another with all flowers that are not Setosa. Then, we can plot it.

We can immediately see that the petal length feature seems to discriminate against the Setosa. Now we’ll only analyse this feature.

Let’s compare the Petal length distribution of Setosa against the all flowers.