# Use the Cars93 dataset from the MASS package to
# Use KNN to classify Origin of cars using cars characteristics
set.seed(8675309)
rm(list = ls())
library(MASS) # Cars93 dataset
library(class) # KNN function
library(gmodels) # CrossTable function

data(Cars93)

# remove incomplete rows
cars <- Cars93[complete.cases(Cars93), ]

# hold out 25 random observations
holdOutIndex <- sample(1:nrow(cars), 25, replace = FALSE)

# get numeric variables and rescale them
numerics <- sapply(cars, is.numeric)
# separate the labels
target <- cars$Origin
trainingTarget <- target[-holdOutIndex]
testingTarget <- target[holdOutIndex]

scaledCars <- data.frame(sapply(cars[numerics],
                                    function(x) scale(x, center = TRUE, scale = TRUE)))

# compare pre and post scaling
summary(cars)
summary(scaledCars)

# build back the training and testing dataset
training <- scaledCars[-holdOutIndex, ]
testing  <- scaledCars[holdOutIndex, ]

# run the model [default to k = sqrt(nobs) ~ 9]
k <- 9
knnPredictions <- knn(train = training, test = testing, cl = trainingTarget, k = k)
CrossTable(testingTarget, knnPredictions, prop.chisq = FALSE)

# 8 True positives, 12 true negatives, 2 false negatives, 3 false positives