#PROJEKTNI ZADATAK - Sanja Bozic
library(corrplot)
library(RColorBrewer)
library(ggplot2)
library(ggpubr)
library(class)
library(Metrics)
library(caret)
library(cluster)
library(fpc)
library(BBmisc)

#citanje csv datoteka
df1<-read.csv("data1.csv",sep=";",dec=",")
df2<-read.csv("data2.csv",sep=";",dec=",")

#brisanje nepotrebnih atributa
df1 <- df1[-1]
df1 <- df1[-1]
df1 <- df1[-1]
df1 <- df1[-20]
df1 <- df1[-2]

#mijenjane krivih naziva
names(df2)[1] <- "avg.height"
names(df2)[4] <- "Height.width"
names(df1)[18] <- "Theta"
names(df2)[18] <- "Theta"

#sortiranje headera u columnama
df1 <- df1[ , order(names(df1))]
df2 <- df2[ , order(names(df2))]

#spajanje dvaju data frame-a
df3 <- rbind(df1,df2)
#str(df3)

#odabir samo glavnih parametara
df4 <- df3[c(4:9,14:16,18)]
attach(df4)

#VIZUALIZACIJA PODATAKA
head(df4,5)

corrplot(cor(df3), type="upper", order="hclust",
         col=brewer.pal(n=8, name="RdYlBu"))

corrplot(cor(df4), type="upper", order="hclust",
         col=brewer.pal(n=8, name="RdYlBu"))

plot_graph <- function(Firmness) {
  ggplot(data=df4, aes(x=Firmness, y=Zs)) +
    geom_point(col="firebrick") +
    geom_smooth(method="lm", se=F)
}

plot_graph(FirmnessAv) + labs(title="Ovisnost izme�u otpora el. struje breskve i prosjeka zrelosti breskve")

f1 <- plot_graph(firmness1)
f2 <- plot_graph(firmness2)
f3 <- plot_graph(firmness3)
f4 <- plot_graph(firmness4)

ggarrange(f1, f2, f3, f4,
          labels = c("A", "B", "C", "D"),
          ncol = 2, nrow = 2)

#SUMIRANJE PODATAKA
#mean
mean_vrijednosti <- c(mean(FirmnessAv),mean(SSC.TA),mean(density.g.cm3),mean(volume.cm3),mean(Theta),mean(Zs))
meanCol_nazivi <- c("FirmnessAV", "SSC.TA", "density.g.cm3","volume.cm3","Theta","Zs")
names(mean_vrijednosti) <- meanCol_nazivi
mean_vrijednosti

#standard dev
sd_vrijednosti <- c(sd(FirmnessAv),sd(SSC.TA),sd(density.g.cm3),sd(volume.cm3),sd(Theta),sd(Zs))
sdCol_nazivi <- c("FirmnessAV", "SSC.TA", "density.g.cm3","volume.cm3","Theta","Zs")
names(sd_vrijednosti) <- sdCol_nazivi
sd_vrijednosti

length_vrijednosti <- length(Zs)
length_vrijednosti

summary(df4)
round(cor(df4),2) #tablica korelacije

#KNN
df4 <- normalize(df4,method="range",range=c(0,1), margin = 1L, on.constant = "quiet")
df4

set.seed(72)
ran <- sample(1:nrow(df4), 0.8 * nrow(df4))
x_train <- df4[ran,]
x_test <- df4[-ran,]

#mi�em stupac sa onim �to predvi�am
x_train <- x_train[-6]
x_test <- x_test[-6]

#ovdje je samo stupac sa onim sta zelim predvidati
y_train <- df4[ran,6]
y_test <- df4[-ran,6]

knn_model <- train(x_train,y_train,method="knn")
knn_model
predictions <- predict(knn_model, x_test)
predictions

mae(y_test,predictions) #mean apsolute error - model je u prosjeku u krivu za 3,59%
mse(y_test,predictions) #mean squared error
rae(y_test,predictions) #relative absolute error
rmse(y_test,predictions) #root mean squared error
mdae(y_test,predictions) #median apsolute error

#proba sa drugim vrstama k
knn.3 <- knn(train=x_train,test=x_test,cl=y_train, k=3)
#pretvorba iz factora u numericki vektor
knn.3 <- as.numeric(as.character(knn.3))
mae(y_test,knn.3) #3,90% mae

#proba sa drugim vrstama k
knn.7 <- knn(train=x_train,test=x_test,cl=y_train, k=7)
#pretvorba iz factora u numericki vektor
knn.7 <- as.numeric(as.character(knn.7))
mae(y_test,knn.7) #5,45% mae

#CLUSTERS
#odre�ivanje broja clustera
wss <- (nrow(df4)-1)*sum(apply(df4,2,var))
for (i in 2:15) wss[i] <- sum(kmeans(df4,
                                     centers=i)$withinss)
plot(1:15, wss, type="b", xlab="Broj clustera",
     ylab="Unutar skupina zbroj kvadrata",
     col="red",
     main="Odre�ivanje broja clustera")

# K-Means Cluster Analysis
fit <- kmeans(df4,3)
fit

# cluster means
aggregate(df4,by=list(fit$cluster),FUN=mean)

# append cluster assignment
data_with_cluster <- data.frame(df4, fit$cluster)
data_with_cluster

#vizualizacija clustera
plotcluster(df4, fit$cluster, xlab="Theta",ylab="") +
  title(main="Prikaz clustera")

#69 u prvom clusteru, 72 u drugom i 59 u trecem