Iris Species prediction using a shiny app

This presentation shows the work done to build a shiny app that predicts Iris Species based on use input that can be found at https://joseo.shinyapps.io/IrisPrediction/.

This application uses sliders as a form of input for the petal/sepal width/length predictors and outputs back a Species prediction based on these inputs. It also includes boxplots of each of the predictors based on species and overlays on them the users' input.

The applicatin also includes a help/usage tab in order to ease their use.

In order to build this application, the following steps have been performed:

• Create a new shiny app using R's menus
• Build a prediction model on the Iris dataset in R
• Save that model to a file inside the Shiny App directory structure
• Adapt the shiny app:
  • To load the model at startup
  • To present the users with a set of sliders to read their inputs
  • To read those inputs, make a prediction and show it to the user, outlining their inputs in the corresponding boxplots

library(dplyr)
library(caret)
data(iris)
inTrain <- createDataPartition(iris$Species, p=0.75, list=FALSE)
trainingData <- iris[inTrain,]
testingData <- iris[-inTrain,]
rfModel <- train(Species ~ ., data=iris,method="rf")
save(rfModel,file="Test/data/model.rda")
predicted <- predict(rfModel,testingData)
mean(predicted==testingData$Species)

[1] 1

We can highlight two important factors on the shiny app

• Loading the model outside the shinyServer function call

  load(file="data/model.rda",.GlobalEnv)
  

• Inside the shinyServer function call, using reactive programming to update our output

  predictionText<- reactive({
      newData <- data.frame(Sepal.Length=input$Sepal.Length,
      ...)
      p <- predict(rfModel,newData)
      levels(iris$Species)[p[1]]
      })