Rob Hindman introduces a forecasting course covering time series visualization, exponential smoothing, ARIMA models, and forecast accuracy. You'll work with Australian spending data to learn prediction intervals that widen as forecasts extend further ahead.
Chapters
Hi, I'm Rob Hindman. I'm a Professor of Statistics at Monash University in Australia and I'll be your instructor for this Data Camp course on forecasting in R. In this course you will learn how to visualize time series in order to discover the useful information you need for forecasting. We'll also consider some very simple forecasting methods, some intermediate level methods such as exponential smoothing and ARIMA models, as well as some more advanced methods. Throughout the course we'll measure how accurate our forecasts are and how to decide which method to use in each case. Everything we cover in this course is discussed in more detail in my textbook with Georgia Thanosopoulos. It is freely available online so you can always refer to it if you need more information. The book uses R throughout and shows the code for almost all graphs and analyses. Prediction is a big topic and in this course we're going to focus on a particular type of prediction, namely forecasting time series. A time series is simply a series of data observed over time. In this course we deal only with regularly spaced time series. For example, the data could be observed every hour, or every day, or every month, every quarter, or every year. Provided the observation intervals are equally spaced. They are called regularly spaced time series. Here is an example of a monthly time series of total expenditure on eating out In Australia. There's a strong trend driven by a mix of population growth and an increase in disposable income, and there's some seasonality. In recent years, eating out costs have peaked in December due to Christmas and end of year events and they drop in February due to it being a short month. Forecasting is the task of estimating how a time series like this will continue into the future. Of course, there's considerable uncertainty associated with such estimates, so we normally also provide an estimate of that uncertainty in the form of prediction intervals. This is an example of what such forecasts might look like. In this example, the forecast values are shown in blue along with 80% prediction intervals and 95% prediction intervals. These indicate how uncertain our forecasts are. The further ahead we forecast, the wider these prediction intervals tend to be. In this course we will discuss various forecasting methods for time series data that take account of trend, seasonality and other features of the data. We will also look at how we can measure the accuracy of these forecasts and how to go about selecting a good forecasting model. That's enough background information. Let's get going with your first interactive exercise.
