Final_MidTerm.R

install.packages("xlsx")
library(xlsx)
library(dplyr)

#Cleaning the dataset
data1 <- read.csv('C:/Users/gadep/OneDrive/Desktop/ADS/CompleteDataset.csv')
data2 <- read.csv('C:/Users/gadep/OneDrive/Desktop/ADS/fifa19.csv')


m=merge(data1,data2,by="ID")

data3=merge(data1,data2,by="ID")
finaldata <- m [c(0:11,14:75,86)]
write.csv(finaldata, file="NewADS.csv")

data3 <- read.csv('C:/Users/gadep/OneDrive/Desktop/ADS/datanew.csv')

#DatatypeIdentification

str(data3)

#Finding the mean age of the Player
table(data$Age)
mean(data$Age)

#Visualizations
install.packages("rsample")
install.packages("randomForest")
install.packages("caret")
install.packages("dplyr")
install.packages("tidyverse", dependencies=TRUE)

library(tidyr)
library(ggplot2)
library(tidyverse)
library(lubridate)

#distribution in histogram based on age
data4 <- read.csv('C:/Users/gadep/OneDrive/Desktop/ADS/DirtyData.csv')
g_age <- ggplot(data = data4, aes(data4$Age))
g_age +
  geom_histogram(col="orange", aes(fill = ..count..)) + ggtitle("Distribution based on Age")

#distribution in histogram based on Value
data4 <- read.csv('C:/Users/gadep/OneDrive/Desktop/ADS/DirtyData.csv')
g_Value <- ggplot(data = data4, aes(Value))
g_Value +
  geom_histogram(col="orange", aes(fill = ..count..)) + ggtitle("Distribution based on Value")

#distribution on Position and Age
data5 <- read.csv('C:/Users/gadep/OneDrive/Desktop/ADS/DirtyData.csv')
g_age <- ggplot(data = data5, aes(data5$Age))
g_age + geom_density(col="orange", aes(fill = Preferred.Positions), alpha=0.5) + facet_grid(.~Preferred.Positions) + ggtitle("Distribution based on Age and Position")

#Graph to show Comparing Market Value of Players with Overall Rating')

ggplot(data3,aes(x=Overall.x,y=Value)) + geom_point(alpha=0.3) + 
  labs(x='Overall Rating',y='Value',title='Comparing Market Value of Players with Overall Rating')

#Divide FIFA 18-19 into training and test datasets (80%/20%)
set.seed(1)
row.number <- sample(1:nrow(data3), 0.8*nrow(data3))
train = data3[row.number,]
test = data3[-row.number,]
dim(train)
dim(test)

# missing values is replaced by Avg of values in that column
for(i in 1:ncol(data3)){
  if(is.numeric(data3[,i])==TRUE)
  {
    data3[is.na(data3[,i]), i] <- mean(data3[,i], na.rm = TRUE)
  }
}

#Removing unneccasary variables
data3<-data3[-c(1:4,6:8,11,12)]
data3<-subset(data3,select=-c(Preferred.Positions))

#removing 1st column
data3<-data3[-c(1)]

#Applying PCA and decision tree on that
pca_train<-data3[(1:nrow(train)),1:65]
pca_train_y<-data3[(1:nrow(train)),66]
pca_test<-data3[-(1:nrow(train)),1:65]
pca_test_y<-data3[-(1:nrow(train)),66]
dim(pca_train)

# Apply PCA
prin_comp<-prcomp(pca_train,scale=T)
str(pca_train)
st_dev<-prin_comp$sdev
pr_var<-st_dev^2
prop_varex <- pr_var/sum(pr_var)

plot(cumsum(prop_varex), xlab = "Principal Component",
     ylab = "Proportion of Variance Explained",
     type = "b")
cumsum(prop_varex)
# BAsed on 95% of explained variance first 12 PCS are selected
train_data<-data.frame(value=pca_train_y,prin_comp$x)
train_data<-train_data[,1:13]

# Modelling using decision tree
model_tree= rpart(value ~ . ,data = train_data)
# transform test to pca
test_data <- predict(prin_comp, newdata = pca_test)
test_data<-as.data.frame(test_data)
test_data<-test_data[,1:12]

# Prediction
rpart_prediction <- predict(model_tree, test_data)
rmse(rpart_prediction,pca_test_y)

#linear regression
LinearRegression = lm(Value ~ . ,data = train )
summary(LinearRegression)

# Checking for correlation
# Correlation heatmap
cormat <- round(cor(data3),2)
cormat
library(reshape2)
melted_cormat <- melt(cormat)

library(ggplot2)
ggplot(data = melted_cormat, aes(x=Var1, y=Var2, fill=value)) +
  geom_tile()

dim(train)
# Removing highly correlated variables
tmp<-cor(train[1:63])
tmp[upper.tri(tmp)] <- 0
diag(tmp)<-0
trainnew <- train[,!apply(tmp,2,function(x) any(x > 0.80))]
str(trainnew)

# Model Creation with uncorrelated variables
LinearRegression = lm(Value ~ . ,data = trainnew )
summary(LinearRegression)
# Still the same very low R square value
pred<- predict(LinearRegression1,test)

library(Metrics)
rmse(pred,test_data$Value)
#rmse(testfit1,test$Value)
annova(LinearRegression)

summary(LinearRegression)

#Accuracy Value
accuracytable = table(test$Value, pred)
accuracy=sum(diag(accuracytable))/sum(accuracytable)*100
accuracy
#-----------------------------------------------------------------------------------------------------
#Decision tree and prediction
#Y1 prediction
library(rpart)
model_tree= rpart(Value ~ . ,data = train)
p= predict(model_tree,test)

#Root Mean Square rror Value
rmse(p,test$Value.y)

#Comparison of the Predicted test values

table(p,test[,2])
p
test[,2]
summary(model_tree)

plot(model_tree,uniform = TRUE,main = "Regression Tree")
text(model_tree,use.n = TRUE , all = TRUE ,cex =.8)

error <- test$Value - p
head(test$Value)
head(p)
head(error)
rmse <- sqrt(mean((error)^2))
library(Metrics)
rmse(test$Value,p)

model_tree= rpart(Value ~ . ,data = test)
p= predict(model_tree,test)

#R square

rr <- train$Value - predict(model_tree)
rr
summary(model_tree)

r_square_rf <- 1 - var(rr)/var(train$Value )
options(digit = 2)
c("R Square is", format(round(r_square_rf,2), nsmall = 2))
# Accuracy
accuracytable = table(test$Value, p)
accuracy=sum(diag(accuracytable))/sum(accuracytable)*100
accuracy
#----------------------------------------------------------------------------------------------------
#Random Forest

install.packages("rsample")
install.packages("randomForest")
install.packages("caret")
library(randomForest)
library("rsample")
library(tidyr)
dim(data3)
names(data3)
class(data3)
str(data3)
summary(data3)
summary(train)

fit.rf <- randomForest(formula = Value ~., data = train)
pred.rf <- predict(fit.rf, test)

rmse.rf <- sqrt(mean(((pred.rf) - test$Value)^2))
(rmse.rf)
c(RMSE = rmse.rf, pseudoR2 = mean(fit.rf$rsq))
plot(pred.rf,test$Value, xlab = "Error", ylab = "Value", pch = 3)

#Random forest prediction
library(Metrics)
rmse(test$Value,p)

summary(fit.rf)

# Accuracy
accuracytable = table(test$Value, pred.rf)
accuracy=sum(diag(accuracytable))/sum(accuracytable)*100
accuracy

#R Square Error
rr <- train$Value - predict(fit.rf)
rr

r_square_rf <- 1 - var(rr)/var(train$Value )
options(digit = 2)
c("R Square is", format(round(r_square_rf,2), nsmall = 2))

#------------------------------------------------------------------------------------------------------------
#xgboost
install.packages("xgboost")
install.packages("Metrics")

library(rJava)
library(xlsx)
library(xgboost)
library(Metrics)

data4 <- read.csv('C:/Users/gadep/OneDrive/Desktop/ADS/NewFifa1.csv')
data <- data.frame(data3)

nrow(data)
set.seed(71)
test_sub <- sample(nrow(data), (1/9)*nrow(data))
train_data <- data[-test_sub,]
test_data <- data[test_sub,]
train_data

train_matrix <- xgb.DMatrix(data = as.matrix(train_data[!names(train_data) %in% c("Value")]), label = train_data$Value)
data.xgb = xgboost(data=train_matrix, max_depth=3, eta = 0.2, nthread=3, nrounds=40, lambda=0
                   , objective="reg:linear")
predict.xgb <- predict(data.xgb, data.matrix(subset(test_data,select=-c(Value))))
summary(predict.xgb)
rmse.xgb <- rmse(as.matrix(test_data["Value"]), as.matrix(predict.xgb))
rmse.xgb