# Modelling and Projecting  of Maternal mortality ratio (per 100 000 live births) in Somalia using 7 Different Models

# Required Packages
library(psych) # Descriptive Analytics
library(tseries) # Diagnostics Analytics 
library(TSstudio) # Data Visualizations and Data partioning
library(forecast) # Prediction 
library(Metrics) # Prescriptive - Model Diagnostics
library(haven)
library(plotly) # More Visualizations
library(forecastHybrid) # For Hybrid Models
library(nnfor) 


# Data Loading and Initial Exploration
Tuberculosis = read.csv(file.choose(),header = T)
head(Tuberculosis)
tail(Tuberculosis)
describe(Tuberculosis)

# Data Transformation
Tuberculosis_Cases_in_Bosaso <-ts(Tuberculosis$TB,start = 1996,
                                  end = 2024, frequency = 1)

# Visualization
windows()
ts.plot(Tuberculosis_Cases_in_Bosaso,xlab="Year",
        ylab="Neonatal_Mortality_Rate",main="Time Series Analysis and Forecasting of Tuberculosis Cases in Bosaso, 
        Somalia (1996–2024)", col="darkblue")

# Splitting Data sets (Test + Train)
splitdata <- ts_split(Tuberculosis_Cases_in_Bosaso, sample.out = 8)
train <- splitdata$train # Ends in 2016
test <- splitdata$test   # 2017-2024
train
test

# Diagnostic Tests
# Test 1  
pp.test(train) #p< 0.05
kpss.test(train) # p> 0.05



# Number of Differencing
ndiffs(train)

#  Differencing 
first<-diff(train)
pp.test(first)
kpss.test(first)


windows()
ts.plot(first,xlab="Year",
        ylab="Tuberculosis_Cases_in_Bosaso ",main="Time Series Analysis and Forecasting of Tuberculosis Cases in Bosaso, 
        Somalia (1996–2024)", col="darkblue")


# Predictive Modelling

# 6 Single Models 

# ARIMA Model + 6 of its Alternatives (ARIMA, TBATS, BATS, Theta, ETS, ARFIMA and ARNN)

# ARIMA (p,d,q) = (p,1,q)
windows()
par(mfrow=c(2,2))
acf(first,main="ACF of Time Series Analysis and Forecasting of Tuberculosis Cases in Bosaso, 
        Somalia (1996–2024)")  # helping for q value selection
pacf(first,main="PACF of Time Series Analysis and Forecasting of Tuberculosis Cases in Bosaso, 
        Somalia (1996–2024) ")  # helping for p value selection

auto.arima(train)      # ARIMA (p,d,q)


# 5. Modelling - Single Models

# 5.1: ARIMA Model
fitARIMA=auto.arima(train)
summary(fitARIMA)
predARIMA=forecast::forecast(fitARIMA, h=8)
print(predARIMA)
windows()
plot(predARIMA)                          
a1<-forecast::accuracy(predARIMA, test);a1
smape(test,predARIMA$mean)
forecast(ets(Tuberculosis_Cases_in_Bosaso),h=7) # 7 Coming years
test_forecast(forecast.obj = predARIMA, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))

# 5.2 ETS Model
fitETS=ets(train)
summary(fitETS)
predETS=forecast::forecast(fitETS, h=8)
print(predETS)
windows()
plot(predETS)
a2<-forecast::accuracy(predETS, test);a2
smape(test,predETS$mean)
forecast(ets(Tuberculosis_Cases_in_Bosaso),h=7) # 7 Coming years
test_forecast(forecast.obj = predETS, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))



# 5.3 TBATS Model
fit_tbats = tbats(train)
summary(fit_tbats)
predTBATS=forecast::forecast(fit_tbats, h=8)
print(predTBATS)
windows()
autoplot(predTBATS)
a4<-forecast::accuracy(predTBATS, test);a4
smape(test,predTBATS$mean)
forecast(tbats(Tuberculosis_Cases_in_Bosaso),h=7) # 7 Years 
test_forecast(forecast.obj = predTBATS, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))

# 5.4 Theta model
fit_theta=thetaf(train, h=8)
print(fit_theta)
windows()
autoplot(fit_theta)
a5<-forecast::accuracy(fit_theta, test);a5
smape(test,fit_theta$mean)
thetaf(Tuberculosis_Cases_in_Bosaso,h=7) # 7 Years
test_forecast(forecast.obj = fit_theta, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))


# 5.5 NNAR model
fit_ARNN = nnetar(train)
predARNN=forecast::forecast(fit_ARNN, h= 8)
print(predARNN)
windows()
plot(predARNN)
a6<-forecast::accuracy(predARNN, test);a6
smape(test, predARNN$mean)
forecast(nnetar(Tuberculosis_Cases_in_Bosaso),h=7) # 7 Years
test_forecast(forecast.obj = predARNN, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))


# 5.6 ARFIMA model
fit_ARFIMA=arfima(train)
predARFIMA = forecast::forecast(fit_ARFIMA, h=8)
print(predARFIMA)
windows()
autoplot(predARFIMA)
a7<-forecast::accuracy(predARFIMA, test);a7
smape(test,predARFIMA$mean)
forecast(arfima(Tuberculosis_Cases_in_Bosaso),h=7) # 7 Years
test_forecast(forecast.obj = predARFIMA, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))


# Prediction Plots for Single Models
windows()
par(mfrow=c(2,3))
plot(predARIMA,col="red")
plot(predETS,col="red")
plot(fit_theta,col="red")
plot(predTBATS,col="red")
plot(predARFIMA,col="red")
plot(predARNN,col="red")


# Decision (accuracy metrics for single models)
a1;a2;a4;a5;a6;

# Hybrid Models

# Hybrid Models: Combined from the provided code and ensuring all possible "8 single time series" combinations are represented.
# The `models` argument in hybrid Model uses single letters: a=ARIMA, e=ETS, f=Theta, n=NNAR, b=BATS, t=TBATS.

# ---------------------------------------------------------
# Two-Model Hybrids (10 Models: a8 - a17)
# ---------------------------------------------------------

# 1. ARIMA + ETS
avg_ae=hybridModel(train, weights="equal",errorMethod = "MASE", models = "ae")
pred_ae = forecast::forecast(avg_ae, h=8)
autoplot(pred_ae)
a8<-forecast::accuracy(pred_ae, test);a8
smape(test, pred_ae$mean)
test_forecast(forecast.obj = pred_ae , actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))

# 2. ARIMA + Theta
avg_af=hybridModel(train, weights="equal",errorMethod = "MASE", models = "af") 
pred_af = forecast::forecast(avg_af, h=8)
autoplot(pred_af)
a9<-forecast::accuracy(pred_af, test);a9
smape(test, pred_af$mean)
test_forecast(forecast.obj = pred_af , actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))




# 3. ARIMA + TBATS
avg_at=hybridModel(train, weights="equal",errorMethod = "MASE", models = "at") 
pred_at = forecast::forecast(avg_at, h=8)
autoplot(pred_at)
a10<-forecast::accuracy(pred_at, test);a10
smape(test, pred_at$mean)
test_forecast(forecast.obj = pred_at , actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))

# 4. ARIMA + NNAR
avg_an=hybridModel(train, weights="equal",errorMethod = "MASE", models = "an") 
pred_an = forecast::forecast(avg_an, h=8)
autoplot(pred_an)
a11<-forecast::accuracy(pred_an, test);a11
smape(test, pred_an$mean)
test_forecast(forecast.obj = pred_an , actual = Tuberculosis_Cases_in_Bosaso, test = test) %>% 
  layout(legend = list(x = 0.1, y = 0.95))

# 5. ETS + Theta
avg_ef = hybridModel(train, weights="equal", errorMethod = "MASE", models = "ef") 
pred_ef = forecast::forecast(avg_ef, h=8)
autoplot(pred_ef)
a12 <- forecast::accuracy(pred_ef, test); a12
smape(test, pred_ef$mean)
test_forecast(forecast.obj = pred_ef, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>%
  layout(legend = list(x = 0.1, y = 0.95))

# 6. ETS + NNAR
avg_en = hybridModel(train, weights="equal", errorMethod = "MASE", models = "en") 
pred_en = forecast::forecast(avg_en, h=8)
autoplot(pred_en)
a13 <- forecast::accuracy(pred_en, test); a13
smape(test, pred_en$mean)
test_forecast(forecast.obj = pred_en, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>%
  layout(legend = list(x = 0.1, y = 0.95))

# 7. ETS + TBATS
avg_et = hybridModel(train, weights="equal", errorMethod = "MASE", models = "et") 
pred_et = forecast::forecast(avg_et, h=8)
autoplot(pred_et)
a14 <- forecast::accuracy(pred_et, test); a14
smape(test, pred_et$mean)
test_forecast(forecast.obj = pred_et, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>%
  layout(legend = list(x = 0.1, y = 0.95))

# 8. Theta + NNAR
avg_fn = hybridModel(train, weights="equal", errorMethod = "MASE", models = "fn") 
pred_fn = forecast::forecast(avg_fn, h=8)
autoplot(pred_fn)
a15 <- forecast::accuracy(pred_fn, test); a15
smape(test, pred_fn$mean)
test_forecast(forecast.obj = pred_fn, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>%
  layout(legend = list(x = 0.1, y = 0.95))

# 9. Theta + TBATS
avg_ft = hybridModel(train, weights="equal", errorMethod = "MASE", models = "ft") 
pred_ft = forecast::forecast(avg_ft, h=8)
autoplot(pred_ft)
a16 <- forecast::accuracy(pred_ft, test); a16
smape(test, pred_ft$mean)
test_forecast(forecast.obj = pred_ft, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>%
  layout(legend = list(x = 0.1, y = 0.95))

# 10. NNAR + TBATS
avg_nt = hybridModel(train, weights="equal", errorMethod = "MASE", models = "nt") 
pred_nt = forecast::forecast(avg_nt, h=8)
autoplot(pred_nt)
a17 <- forecast::accuracy(pred_nt, test); a17
smape(test, pred_nt$mean)
test_forecast(forecast.obj = pred_nt, actual = Tuberculosis_Cases_in_Bosaso, test = test) %>%
  layout(legend = list(x = 0.1, y = 0.95))