Sign in or start a free trial to avail of this feature.
1. Introduction to Time Series Forecasting
In this lesson, we introduce the concept of a time series forecast. We will look at the differences between time series analysis and regression analysis.
Time Series Forecasting
- Time series data is any single variable that has been observed over a period and indexed in time order
- Forecasting this data depends on a variety of factors, with historic data playing a very large role
- Time series forecasting has some crossover with the concept of regression, but only requires a single set of data where regressions find the line of best fit between multiple factors
Time series data is any single variable that has been observed over a period, indexed in time order. This could be the height of the tide, the Euro-Dollar exchange rate, or the number of hits on a website landing page. The Alteryx times series tools analyze such periodic data with the goal of forecasting future values. The accuracy of these forecasts depends on a number factors not least how much historic data is available and how completely we understand what influences the results. For this reason we're often less concerned with the forecast itself, rather more concerned with the degree of confidence about our forecast. For example, we can forecast the time of high tide tomorrow with a high degree of accuracy but forecasting the Euro-Dollar exchange rate this time tomorrow is more of a coin toss. However, we can be much more confident that the Euro-Dollar exchange rate will be plus or minus three percent of where it is today. Over the course of the next number of lessons we're going to introduce you to some of the most popular forecasting techniques and how they're best deployed. As we run through our examples you should bear in mind that forecasting is often an iterative process. Even the best forecasts are unlikely to predict the future precisely by they can give you an indication as what results are more likely. Time series forecasting takes historic data and uses a data model to predict future values. In a previous course we looked at regression analysis. While a regression seeks to find the line of best fit between multiple factors, in its simplest form a time series forecast only requires a single set of data together with the corresponding dates or times. For example, consider the online marketing dataset we've previously used. We can see that this dataset contains information on daily sales, marketing spend by channel and several other factors. In our statistics lessons we used the marketing spend by channel on a particular day to predict that day's sales. Instead the times series forecast looks at the sales each day and attempts to forecast future sales. There is of course a degree of cross over between these two concepts. When forecasting future sales the model assumes that the factors that influence past sales will be consistent in the future. If marketing spend is significantly changed for a certain channel this will likely alter future sales. Sophisticated models can account for changes like this, however, we'll concentrate on the more straightforward techniques. We'll stop here for now and look into the different forecast methodologies available in Alteryx in the next lesson.