Forecasting
There are a lot of different ways for forecasting; depending on your case you use them. The library that you would need first to include in R is “library(forecast)”. There are ETS, TBATS, NNETAR, CROSTON, SNAIVE, NAIVE, RWF, Arima, and lots of others. Each of them uses different algorithms for prediction and of course, you can always change the variables of them to change the model and get a different result. The “Arima” is one of the best ways to forecast and plot the result. But You should know that all of these methods get a time series as an input!
For creating a time series out of your data you can use “ts()” as your converter to time series. Of course “ts()” has some variables that you should know about to work with.
The “ets()” has different models such as “AAA”, “MMA”, “MMN”, and others. Each of them gives you different results depending on your data. You might want to study that to figure out which one is better for your case to use.
The “tbats()” also have different models and variables that you should know about. This method usually used when you have a pattern changes seasonally.
The “NNETAR()” is a neural network method for the prediction that also has different variables that you should know about.
Usually, when you get the result of each method you use, then you can pass it to forecast() function to forecast the data for you.
ETS <- forecast(ets(ts[,1], model="AAA", damped=TRUE), biasadj=TRUE,h=30) ARIMA <- forecast(auto.arima(ts[,1], lambda=0, biasadj=TRUE), h=30) TBATS <- forecast(tbats(ts[,1], lambda=0), biasadj=TRUE, h=30)
Note: “h=30” is actually the number of forecast data you wanted to be.
One method that I really like and it is cool, is the combination of different methods. Like you get the average of all different results. In many cases works perfectly fine!
Combination <- (ETS_cases[["mean"]] + ARIMA_cases[["mean"]] + TBATS_cases[["mean"]]) /3
So, you get the “mean” of each method you used and sum them up then divide them by the number of methods you have.
The Correct Way To Start
If you know a way that works for your case and you are sure about it, go for it. But the best way to figure out what exactly is the best way to forecast/predict the data you have is to plot it in a different way first and see what is the pattern of your data and find a relationship between time or another dependent metric/s.
First, you might need to include this library in R “library(ggplot2)”. The “ggplot” is one of the easiest ways to plot predicted data and work within forecast scenarios.
You might want to start with looking at your data via “geom_line()” or “geom_point()” and then “boxplot()”, you can also use “scatter.smooth()” mostly if you are doing Linear regression with “lm()”, and then maybe “checkresiduals()”.
These are different ways of visualizing your data and analyzing which method would work the best for your data.
Plot and ggplot are easiest way to plot the data.
You can do plot(mydata) which shows you your data or you can do ggplot() + geom_line() + geom_smooth() + geom_smooth().
ggplot() +
geom_line(data = ts, aes(x=ts$date,
y=ts$cases) ) +
geom_smooth(data=forecs, aes(x=forecs$date,
y=forecs$cases), color="green" ) +
geom_smooth(data=forecs, aes(x=forecs$date,
y=forecs$age), color="red" ) +
geom_smooth(data=forecs, aes(x=forecs$date,
y=forecs$health), color="blue" ) +
labs(title="Forecasting Cases ",
subtitle="From Vietnam",
y="Number of cases", x="Date")The above code is a way to plot different forecast line in a single plot along with the original data.
You can see the result below:
Another way to plot the forecasting data is the “autoplot()”. It is another way to see the three or more different forecast results in one plot.
ETS <- forecast(ets(ts[,1], model="MMN", damped=FALSE), biasadj=TRUE, h=30)
ARIMA <- forecast(auto.arima(ts[,1], lambda=0, biasadj=TRUE), h=30)
TBATS <- forecast(tbats(ts[,1], lambda=0), biasadj=TRUE, h=30)
Combination <- (ETS[["mean"]] + ARIMA[["mean"]] +
TBATS[["mean"]])/3
autoplot(ts) +
autolayer(ETS, color="blue", PI=FALSE) +
autolayer(ARIMA, color="red", PI=FALSE) +
autolayer(Combination, color="green") +
xlab("Months") + ylab("Every 2 weeks Cases") +
ggtitle("Forecasting Cases in Brazil")