autoput <- read.csv("travel-times.csv",stringsAsFactors = FALSE)
str(autoput)
apply(X=autoput,
      MARGIN=2,
      FUN=function(x){
        length(which(is.na(x)))

      })
apply(X=autoput,
      MARGIN=2,
      FUN=function(x){
        length(which(x=="" | x=="-"))

      })
#Izbacujemo varijablu comments jer ima mnogo nedostajucih vrednosti, a i nije
#bitna za analizu.
table(autoput$GoingTo)
autoput$GoingTo[autoput$GoingTo=="" | autoput$GoingTo=="-"] <- "Work"
#autoput$FuelEconomy[autoput$FuelEconomy=="" | autoput$FuelEconomy=="-"] <- NA
autoput$FuelEconomy <- as.numeric(autoput$FuelEconomy)
shapiro.test(autoput$FuelEconomy)
medijana <- median(autoput$FuelEconomy,na.rm=T)
autoput$FuelEconomy[is.na(autoput$FuelEconomy)] <- medijana

str(autoput)
autoput$Date <- NULL
autoput$StartTime <- NULL
autoput$DayOfWeek <- as.factor(autoput$DayOfWeek)
autoput$GoingTo <- as.factor(autoput$GoingTo)
str(autoput)

percent <- as.numeric(quantile(autoput$Congestion407,0.6))


autoput$Take407All <- ifelse(test=(autoput$Congestion407<percent & autoput$Comments==""),
                             yes="YES",
                             no="NO")
autoput$Take407All <- as.factor(autoput$Take407All)
head(autoput$Take407All)
autoput$Congestion407 <- NULL
autoput$Comments <- NULL
#trening i test set
library(caret)
set.seed(10)
indexes <- createDataPartition(autoput$Take407All,p=0.8,list=FALSE)
train.data <- autoput[indexes,]
test.data <- autoput[-indexes,]


library(e1071)
library(caret)
set.seed(10)
numFolds = trainControl(method = "cv", number = 10) # define <k>-fold cross-validation parameters
cpGrid= expand.grid(.k = seq(from = 3, to = 25, by = 2))
cross.val <- train(Take407All ~.,
                    data = train.data,
                     method = "knn",
                    trControl = numFolds, tuneGrid = cpGrid)
cross.val
plot(cross.val)
#Dobili smo da je optimalna vrednost za k=23;
#model
str(autoput)
library(class)
knn1 <- knn(train=train.data[,-10],
           test=test.data[,-10],
           cl=train.data$Take407All,
           k=23)
knn1
table(knn1)
knn1.cm <- table(true=test.data$Take407All,predicted=knn1)
table(train.data$Take407All)
str(train.data$GoingTo)