Predicting Iris Species Using K-Nearest Neighbors Algorithm

• The iris dataset consists of 150 observations of three different species of iris (virginica, setosa and versicolor). Each observation includes data for Petal Length, Petal Width, Sepal Length and Sepal Width.
• We use a k-nearest neighbors algorithm (k=13) to classify an unknown species type based upon the length and width measurements above.

• Finally, we build a Shiny app which allow users to interactively predict the species based on these measurement variables.

• This example inspired by Jalayer Academy's tutorial video on Youtube: link

• Our Shiny app is deployed on RPubs at: THIS LINK
• It consists of ui.R and server.R files
  • ui.R creates the user interface including the four input measurements for petal and sepal
  • server.R provides reactive functions which take the four input measurements as inputs and immediately provide a species prediction using a k-nearest neighbors classification algorithm
• For higher accuracy, our algorithm nomalizes the predictor variables using the below function:

  normalize <- function(x) {
    return( (x - min(x)) / (max(x) - min(x)) ) 
  }

• Below is the knn() algorithm inside the reactive function:

library(class)
library(shiny)
  m1pred <- reactive({
    Sepal.Length.Input <- input$sliderSepalLength
    Sepal.Width.Input <- input$sliderSepalWidth
    Petal.Length.Input <- input$sliderPetalLength
    Petal.Width.Input <- input$sliderPetalWidth
    knn(train = iris_train, test = data.frame(
      (Sepal.Length.Input - min(iris$Sepal.Length)) / (max(iris$Sepal.Length) - min(iris$Sepal.Length)),
      (Sepal.Width.Input - min(iris$Sepal.Width)) / (max(iris$Sepal.Width) - min(iris$Sepal.Width)),
      (Petal.Length.Input - min(iris$Petal.Length)) / (max(iris$Petal.Length) - min(iris$Petal.Length)),
      (Petal.Width.Input - min(iris$Petal.Width)) / (max(iris$Petal.Width) - min(iris$Petal.Width))
      ), 
          cl = iris_train_target, k = 13
    )
  })

• This Shiny App demonstrates how we can pair classification algorithms with simple user interfaces to easily create accurate predictions about species type for irises.
• Classification algorithms have huge business value, e.g. detection of credit card fraud or spam email.
• For significantly more complex algorithms and and larger datasets we will want to integrate our app with some cloud service which can auto-scale and reduce trainig times.