574_Project_v2.Rmd

---
title: "574 Project"
author: "Malvika Mohan"
date: "3/3/2020"
output: pdf_document
---

```{r setup, include=FALSE}
library(tidyverse)
library(MASS)
library(e1071)
library(corrplot)
library(stats)
library(caret)
library(pROC)
```



```{r}
weather_data <- read.csv('C:/RP/datasets/weatherAUS.csv')
weather_data <- data.frame(weather_data)
weather_data <- na.omit(weather_data)
rownames(weather_data) <- 1:nrow(weather_data)
View(weather_data)
weather_data

```



```{r }
#Checking correlation
temp <- weather_data$RainToday
weather_data2 <- weather_data %>% select_if(is.numeric) %>% dplyr::select(-RISK_MM) 

bivrel <- cor(weather_data2,y=NULL,use = "everything",method = "pearson")
corrplot(bivrel, method = "square")

```

```{r }
#Determining the rainfall,average temperature,Humidity trends across three cities =1) Melbourne  


weather_dataMel <-weather_data %>% filter(Location=='Melbourne') 


dev.new()  # Generates a new plot window.

#Plotting rainfall in inches for melbourne
plot(ts (weather_dataMel$Rainfall, start=c(2007), end=c(2017), frequency=1),xlab="",ylab="Rainfall in inches",col=" Blue",main = "Yearly precipitation trends of Melbourne inches")




weather_dataMel$Avgtemp <- weather_dataMel %>% group_by(Date) %>% summarise(avgt=mean(MinTemp,MaxTemp,na.rm=TRUE))

#Plotting average temperature in celsius in Melbourne
plot(ts (weather_dataMel$Avgtemp$avgt, start=c(2007), end=c(2017), frequency=1),ylab="Average temperature in celsius",col="Red",main="Yearly temperature trends of Melbourne ")


#Plotting Humidity percentage at 3pm  in Melbourne
plot(ts (weather_dataMel$Humidity3pm, start=c(2007), end=c(2017), frequency=1),ylab="Percentage of humidity in Melbourne",col="green",main = "Yearly Humidity of Melbourne")

#Plotting Sunshine in Melbourne
plot(ts (weather_dataMel$Sunshine, start=c(2007), end=c(2017), frequency=1),ylab="Number of hours of sunshine per day",col="orange",main = "Number of days of sunshine in Melbourne")
```

```{r }
#Determining the rainfall,average temperature trends across three cities =2) Sydney  

weather_dataSyd <- weather_data %>% filter(Location=='Sydney')


plot(ts (weather_dataSyd$Rainfall, start=c(2007), end=c(2017), frequency=1),xlab="",ylab= "Rainfall in inches",col="Blue",main = "Yearly precipitation trends of Sydeny ")

weather_dataSyd<-na.omit(weather_dataSyd);
weather_dataSyd$Avgtemp <- weather_dataSyd %>% group_by(Date) %>%  summarise(avgt2=mean(MinTemp,MaxTemp,na.rm = TRUE))



#Plotting average temperature in celsius in Sydney
plot(ts (weather_dataSyd$Avgtemp$avgt2,  start=c(2007), end=c(2017), frequency=1),xlab="",ylab="Average temperature in celsius",col="Red",main="Yearly temperature trends of Sydney")

#Plotting Humidity percentage at 3pm in Sydney
plot(ts (weather_dataSyd$Humidity3pm, start=c(2007), end=c(2017), frequency=1),ylab="Percentage of humidity at 3pm in  Sydney",col=" green",main = "Yearly Humidity percentage of Sydney")


#Plotting Sunshine in Sydney
plot(ts (weather_dataSyd$Sunshine, start=c(2007), end=c(2017), frequency=1),ylab="Number of hours of sunshine per day",col="orange",main = "Yearly Number of hours of sunshine in Sydney")

```

```{r }
#Determining the rainfall,average temperature trends across three cities = 3)Cranberra 
weather_dataCanberra <- weather_data %>% filter(Location=='Canberra')




plot(ts (weather_dataCanberra$Rainfall, start=c(2007), end=c(2017), frequency=1),xlab="",ylab="Yearly precipitation trends of Cranberra inches",col="Blue")


weather_dataCanberra$Avgtemp <- weather_dataCanberra %>% group_by(Date) %>%  summarise(avgt3=mean(MinTemp,MaxTemp,na.rm = TRUE))




#Plotting average temperature in celsius in Cranberra
plot(ts (weather_dataCanberra$Avgtemp$avgt3,  start=c(2007), end=c(2017), frequency=1),xlab="",ylab="Average temperature of Cranberra in degrees",col="Red")


#Plotting percentage of Humidity at 3pm in Canberra
plot(ts (weather_dataCanberra$Humidity3pm, start=c(2007), end=c(2017), frequency=1),ylab="Percentage of humidity at 3pm in Cranberra",col=" green",main = "Yearly Humidity percentage of Cranberra")


#Plotting Sunshine in Canberra
plot(ts (weather_dataCanberra$Sunshine, start=c(2007), end=c(2017), frequency=1),ylab="Number of hours of sunshine",col="orange",main = "Number of days of sunshine in Canberra")

```

```{r }
#Checking for significant variables for a linear regression model.
lower_regression_limit <- lm(Rainfall~1,data =head(weather_data2,1000))
upper_regression_limit <- lm(Rainfall~.,data=head(weather_data2,1000))
step_model <- stepAIC(lower_regression_limit , scope = list(lower = lower_regression_limit , upper = upper_regression_limit), direction ="both",trace = FALSE)
summary(step_model)
```

Retaining the variables as determined by least-squares inference.


```{r }
#Setting up the data for classification.
weather_data3 <- weather_data2 %>% dplyr::select(Humidity9am,Sunshine,WindSpeed9am,Temp3pm,MaxTemp,WindGustSpeed) %>% cbind(Rain_today = temp)

set.seed(32)
split_ind <- sample(nrow(weather_data3),size = 0.75 * nrow(weather_data3),replace = FALSE)
train <- weather_data3[split_ind,]
test <- weather_data3[-split_ind,]
train
attach(train)
```

```{r}
#Logistic regression.
model_logistic <- glm(Rain_today ~.,family = binomial(link = 'logit'),data = train)
summary(model_logistic)
predict_logistic <- predict(model_logistic,test,type = 'response')


#Confusion Matrix at a threshold of 0.5
op_logistic <- data.frame(test$Rain_today, predict_logistic)
op_logistic$classification <- ifelse(op_logistic$predict_logistic > 0.5,1,0)
op_logistic$test.Rain_today <- if_else(op_logistic$test.Rain_today == 'Yes',1,0)
confusionMatrix(factor(op_logistic$classification),factor(op_logistic$test.Rain_today))

#ROC 
par(pty = 's')
roc(train$Rain_today,model_logistic$fitted.values,plot = TRUE,legacy.axes= TRUE,col = '#ff5252',lwd = 4,print.auc = TRUE)

```

```{r}

model_svm <- svm(Rain_today ~. ,train)
predict_svm <- predict(model_svm, test,type = 'response')
summary(model_svm)
model_svm$fitted

#ConfusionMatrix
op_svm <- data.frame(test$Rain_today, predict_svm)
confusionMatrix(factor(op_svm$predict_svm),factor(op_svm$test.Rain_today))


#svm_tune <- tune(svm, y~x, data = data, ranges=list(epsilon=seq(0,1,0.1), cost=seq(1,10,1)))
#ROC
par(pty = 's')
roc(train$Rain_today,if_else(model_svm$fitted == 'Yes',1,0),plot = TRUE,legacy.axes= TRUE,col = '#ff5252',lwd = 4,print.auc = TRUE)
```

```{r}
par(pty = 's')

#Purple curve: logistic regression model 
roc(train$Rain_today,model_logistic$fitted.values,plot = TRUE,legacy.axes= TRUE,col = '#9c88ff',lwd = 4,print.auc = TRUE,print.auc.y = 0.3)


#Green curve: SVM
roc(train$Rain_today,if_else(model_svm$fitted == 'Yes',1,0),plot = TRUE,legacy.axes= TRUE,col = '#4cd137',lwd = 4,print.auc = TRUE,add = TRUE,print.auc.y = 0.2)
```