sales <- c(18,33,41,7,34,35,24,25,24,21,25,20,22,31,40,29,25,21,22,54,31,25,26,35)

tsales <- ts(sales, start = c(2003,1),frequency = 12 ) #numerical values have default plot of scatter
plot
plot(tsales)
plot(tsales, xlab= "time", ylab="monthly sales", main= "sales(2003-2004)")
start(tsales)
end(tsales)
frequency(tsales)
tsales.subset <- window(tsales, start = c(2003,5),end = c(2004,6))#subset the series
plot(tsales.subset)

rm(list = ls()) #cleaning of environment

#Ctrl+L to clear the console

library(forecast)
plot(Nile, main = "raw time series")
plot(ma(Nile, 3), main ="Simple Moving Average (k = 3)")
plot(ma(Nile, 7), main ="Simple Moving Average (k = 7)")
plot(ma(Nile, 15), main ="Simple Moving Average (k = 15)")
lim <- c(min(Nile), max(Nile))
par(mfrow = c(2,2))
plot(Nile, main = "raw time series")
plot(ma(Nile, 3), main ="Simple Moving Average (k = 3)", ylim = lim)
plot(ma(Nile, 7), main ="Simple Moving Average (k = 7)", ylim = lim)
plot(ma(Nile, 15), main ="Simple Moving Average (k = 15)", ylim = lim)
par(mfrow = c(1,1))

#seasonal decomposition
par(mfrow = c(2,1))
plot(AirPassengers)
lair <- log(AirPassengers) #to convert from multiplicative to additive
plot(lair,ylab = "logAirPassengers")
fit <- stl(lair, s.window = "period") #used for decomposition into seasonality, trend, irregularity or
error
plot(fit)
fit$time.series
exp(fit$time.series)#anti log
monthplot(AirPassengers, xlab="",ylab="")
seasonplot(AirPassengers, year.labels = "true", main = "")

#simple exponential
plot(nhtemp)
fit <- ets(nhtemp, model = "ANN")
fit
pred <- forecast(fit,1)
pred
plot(pred,xlab = "year", ylab = "temp", main = " ")
accuracy(fit)

#level 3 exponentail sm
fit1 <- ets(lair , model = "AAA")
fit1
accuracy(fit1)
pred1 <- forecast(fit1,5)
plot(pred1, xlab = "year", ylab = "temp", main = "")
fit2 <- ets(lair)
fit2
accuracy(fit2)
accuracy(fit1)

data_input <- read.csv('C:/Users/Shreya/Downloads/soft.csv')

tsales <- ts(data_input$Sales, start = c(1997,1),frequency = 4)
tsales
plot(tsales)
plot(tsales, xlab= "time", ylab="Quarterly sales", main= "sales(1997-2012)")
par(mfrow = c(1,1))
fit <- stl(tsales, s.window = "period")
fit
plot(fit)
lim <- c(min(tsales), max(tsales))
par(mfrow = c(2,2))
plot(tsales, main = "raw time series")
plot(ma(tsales, 3), main ="Simple Moving Average (k = 3)", ylim = lim)
plot(ma(tsales, 7), main ="Simple Moving Average (k = 7)", ylim = lim)
plot(ma(tsales, 15), main ="Simple Moving Average (k = 15)", ylim = lim)

fit <- ets(tsales, model = "ANN")

fit
pred <- forecast(fit,1)
pred
plot(pred,xlab = "year", ylab = "Sales", main = " ")
accuracy(fit)

fit <- ets(tsales, model = "AAN")

fit
pred <- forecast(fit,1)
pred
plot(pred,xlab = "year", ylab = "Sales", main = " ")
accuracy(fit)

fit <- ets(tsales, model = "AAA")

fit
pred <- forecast(fit,1)
pred
plot(pred,xlab = "year", ylab = "Sales", main = " ")
accuracy(fit)
sales <- read.csv("Soft Drink Sales.csv", header = TRUE, stringsAsFactors = TRUE)
head(sales) #first 6 observations
sales_ts <- ts(sales$Sales, start = c(1997, 1), frequency = 4)
sales_ts
plot(sales_ts)
library(forecast)

############ Linear Trend (Additive) ############

fit_lin_trend <- tslm(sales_ts ~ trend) #we use lm for simple regression but for forecasting we
use tslm-time series linear model
pred <- forecast(fit_lin_trend, h=4)
pred
summary(fit_lin_trend)
accuracy(fit_lin_trend) # we will look for RMSE and MAPE the least value in every model applied will
the best model
plot(pred, xlab = "Time", ylab = "Sales ($)", main = "Actual and forecasted values (Linear Trend)")

##########
par(mfrow = c(2,1))
plot(sales_ts)
lines(fit_lin_trend$fitted, col = "red", lty = 2)
points(pred$mean, col = "blue", pch = 16)
plot(fit_lin_trend$residuals)
par(mfrow = c(1,1))

####### Exponential Trend (Multiplicative) ########

fit_exp_trend <- tslm(sales_ts ~ trend, lambda = 0) #lambda = 0 conversion of Y into log of Y
pred <- forecast(fit_exp_trend, h=4)
pred
accuracy(fit_exp_trend)
plot(pred, xlab = "Time", ylab = "Sales ($)", main = "Actual and forecasted values")

##########
par(mfrow = c(2,1))
plot(sales_ts)
lines(fit_exp_trend$fitted, col = "red", lty = 2)
points(pred$mean, col = "blue", pch = 16)
plot(fit_exp_trend$residuals)
par(mfrow = c(1,1))
__________________________________________________________________________________
############# Quadratic Trend (Amtrak Data) #############
#setwd("D:/Shivani 2017/LBS classes/Data Science 1/Data
Science_2017/Forecasting/Regression_Forecasting")
#when data is polynomial we sqaure or cube the independent variable
ride <- read.csv("Amtrak data.csv", header = TRUE, stringsAsFactors = TRUE)
head(ride)
ride_ts <- ts(ride$Ridership, start = c(1991,1), end = c(2004, 3), freq = 12)
ride_ts
plot(ride_ts, xlab = "Time", ylab = "Ridership", ylim = c(1300, 2300))
fit_quad_trend <- tslm(ride_ts ~ trend + I(trend^2))

pred <- forecast(fit_quad_trend, h=12)

pred

accuracy(fit_quad_trend)
plot(pred, xlab = "Time", ylab = "Sales ($)", main = "Actual and forecasted values")

##########

par(mfrow = c(2,1))
plot(ride_ts)
lines(fit_quad_trend$fitted, col = "red", lty = 2)
points(pred$mean, col = "blue", pch = 16)

plot(fit_quad_trend$residuals)

par(mfrow = c(1,1))
##################################

rm(list = ls())

##########################
##Model with seasonality ### seasonality was not included till, now we include seasonality
fit_season <- tslm(ride_ts ~ season)

pred <- forecast(fit_season, h=12)

pred

accuracy(fit_season)

plot(pred, xlab = "Time", ylab = "Sales ($)", main = "Actual and forecasted values")

######################################

par(mfrow = c(2,1))
plot(ride_ts)
lines(fit_season$fitted, col = "red", lty = 2)
points(pred$mean, col = "blue", pch = 16)

plot(fit_season$residuals)

par(mfrow = c(1,1))

##################################
## Model with trend and seasonality ####
fit_season_trend <- tslm(ride_ts ~ trend + I(trend^2) + season)

pred <- forecast(fit_season_trend, h=12)

pred

accuracy(fit_season_trend)

plot(pred, xlab = "Time", ylab = "Sales ($)", main = "Actual and forecasted values")

#######################################

par(mfrow = c(2,1))
plot(ride_ts)
lines(fit_season_trend$fitted, col = "red", lty = 2)
points(pred$mean, col = "blue", pch = 16)

plot(fit_season_trend$residuals)

par(mfrow = c(1,1))

rm(list = ls())

______________________________________________________________________________
#setwd("D:/Shivani 2017/LBS classes/Data Science/Forecasting/Regression_Forecasting")
ride <- read.csv("Amtrak data.csv", header = TRUE, stringsAsFactors = TRUE)
head(ride)
ride_ts <- ts(ride$Ridership, start = c(1991,1), end = c(2004, 3), freq = 12)
ride_ts
plot(ride_ts, xlab = "Time", ylab = "Ridership", ylim = c(1300, 2300))
library(forecast)

#### LINEAR TREND & SEASONALITY ###

fit_tren_sea <- tslm(ride_ts ~ season)

pred <- forecast(fit_tren_sea, h=12)
pred
summary(fit_tren_sea)

accuracy(fit_tren_sea)
plot(pred, xlab = "Time", ylab = "Sales ($)", main = "Actual and forecasted values (Linear Trend with
Season)")

##########

par(mfrow = c(2,1))
plot(ride_ts, xlim = c(1990, 2006))
lines(fit_tren_sea$fitted, col = "red", lty = 2)
points(pred$mean, col = "blue", pch = 16)

plot(fit_tren_sea$residuals)

par(mfrow = c(1,1))

####### QUADRATIC TREND & SEASONALITY ##########

fit_quad_sea <- tslm(ride_ts ~ trend + I(trend^2) + season)

pred <- forecast(fit_quad_sea, h=12)
pred
summary(fit_quad_sea)
accuracy(fit_quad_sea)
plot(pred, xlab = "Time", ylab = "Sales ($)", main = "Actual and forecasted values (Quadratic Trend
with Season)")

##########

par(mfrow = c(2,1))
plot(ride_ts, xlim = c(1990, 2006))
lines(fit_quad_sea$fitted, col = "red", lty = 2)
points(pred$mean, col = "blue", pch = 16)

plot(fit_quad_sea$residuals)

par(mfrow = c(1,1))

rm(list = ls())
__________________________________________________________________________________
Revised Partition
#setwd("D:/Shivani 2017/LBS classes/Data Science/Forecasting/Regression_Forecasting")

ride <- read.csv("Amtrak data.csv", header = TRUE, stringsAsFactors = TRUE)

head(ride)

ride_ts <- ts(ride$Ridership, start = c(1991,1), end = c(2004, 3), freq = 12)
ride_ts

plot(ride_ts, xlab = "Time", ylab = "Ridership", ylim = c(1300, 2300))

library(forecast)

#### LINEAR TREND & SEASONALITY ###

fit_tren_sea <- tslm(ride_ts ~ season)

pred <- forecast(fit_tren_sea, h=12)
pred
summary(fit_tren_sea)

accuracy(fit_tren_sea)
plot(pred, xlab = "Time", ylab = "Sales ($)", main = "Actual and forecasted values (Linear Trend with
Season)")

##########

par(mfrow = c(2,1))
plot(ride_ts, xlim = c(1990, 2006))
lines(fit_tren_sea$fitted, col = "red", lty = 2)
points(pred$mean, col = "blue", pch = 16)

plot(fit_tren_sea$residuals)

par(mfrow = c(1,1))

####### QUADRATIC TREND & SEASONALITY ##########

fit_quad_sea <- tslm(ride_ts ~ trend + I(trend^2) + season)

pred <- forecast(fit_quad_sea, h=12)
pred

summary(fit_quad_sea)
accuracy(fit_quad_sea)

plot(pred, xlab = "Time", ylab = "Sales ($)", main = "Actual and forecasted values (Quadratic Trend
with Season)")

##########

par(mfrow = c(2,1))
plot(ride_ts, xlim = c(1990, 2006))
lines(fit_quad_sea$fitted, col = "red", lty = 2)
points(pred$mean, col = "blue", pch = 16)

plot(fit_quad_sea$residuals)

par(mfrow = c(1,1))

rm(list = ls())

--------------------------------------------------------------------------------------------------------------------------------------
ARIMA
install.packages("forecast")

library(forecast)
install.packages("tseries")
library(tseries)

### 1. Ensure that the time series is stationary ###

plot the time series to assess the stationarity

plot(Nile)

# Difference the time series to check for the trend

ndiffs(Nile)

# the series is differenced once (lag = 1 is default) and saved as dNile

dNile <- diff(Nile)

 # plot the differenced time series

plot(dNile)

# Apply ADF test to check for stationarity.

adf.test(dNile)
# as now we now we need to difference one timeso we will apply adf to dnile not nile

# ADF test proves that time series is stationary, therefore, proceed further.

### 2. Identifying one or more reasonable models

# possible models are selected based on ACF and PACF plots

par(mfrow = c(2,1))

Acf(dNile)
Pacf(dNile)

par(mfrow = c(1,1))

### 3. Fitting the Models ###

fit <- arima(Nile, order = c(0,1,1))

fit

accuracy(fit)

### 4. Evaluating Model Fit ###

qqnorm(fit$residuals)
qqline(fit$residuals)

Box.test(fit$residuals, type = "Ljung-Box") # box.test value should be >.05 to satisy the test

### 5. Making Forecast ###

forecast(fit, 3)

plot(forecast(fit, 3), xlab = "Year", ylab = "Annual Flow")

####### Automated ARIMA Forecasting ########

fit1 <- auto.arima(Nile)
fit1

forecast(fit1, 3)
accuracy(fit1)

plot(forecast(fit1, 3))
rm(list = ls())