Tableau is one of the favorite data visualization tools for many analytics practitioners. Version 8, the most recent release of Tableau contains a forecasting feature which makes it even more interesting for forecasters who currently use Tableau only as a frontend for their models. For example, we have several customer cases where we built forecasting models using open source tools such as R or RapidMiner, and then created a dashboard using Tableau, so that our customers can have a nice interface when they want to visualize the model results (and all the basic analytics which Tableau is already well known for).
Does this new time series forecasting feature of Tableau mean we now have a single tool for all our forecasting analytics?
This is a basic rundown of what the new forecasting feature can and cannot do. As with all things in Tableau, getting from data to a visual is a breeze. So here is how you would generate forecasts using version 8.
- Connect to your data: I tried both csv and xlsx files. Your data needs to contain at least two columns, out of which one should be a "date" format column. If it is a text (csv) file, make sure that you convert the date column to any of the standard date formats (mm-dd-yyyy, D-M-YY etc).
- Drag and drop your Measure into the "Rows" field and your Dimension (in this case, "Date") into the "Columns" field. As usual Tableau will detect the type of data and automatically select a line plot.
- Right click in the charting area and select Forecast -> Show Forecast from the drop down OR
- From the top menu bar, select Analysis -> Forecast -> Show Forecast
All the things you love about Tableau are nicely brought together in this new feature. It is a snap to generate a forecast using any of the data you already may have (as long as it is in the proper, "forecastable" format described above).
Viewing the model summary - which is done by a call to the summary() function in R for example, is also pretty easy. Analysis->Forecast->Describe Forecast... generates a nice table which contains all the summary info about the data. Model information such as quality metrics (RMSE, MAE, MASE, MAPE, and AIC, which I suspect stands for Akaike Information Criterion) and smoothing parameters, alpha, beta and gamma that were fitted for the data are displayed.
By choosing Analysis->Forecast->Forecast Options ... one can change the type of model. However, here there is not a lot of choice and Tableau has ways to go before it can catch up with the likes of R packages in providing modeling options.
The best part about using the forecasting feature is the easy extensibility. Once you select a particular model for a dataset, then simply dragging and dropping new measures on to the rows field will automatically apply the same forecasting model to the new measures, rapidly in typical Tableau fashion.
As mentioned earlier, the range of modeling choices is limited to one: exponential smoothing. Within this option, you can choose an Automatic, Automatic without seasonality, Trend only, Season only, Trend and Season, no trend or season. For the dataset I played with, the Automatic, Automatic w/o season and no trend or season generated the same result: a simple averaging as seen in the top image above. When I changed the forecast model option to Trend and Season, the forecast results improved greatly and i was able to more or less replicate the results that were obtained earlier using R.
How do the results compare between the forecasting models generated by Tableau and R? I found that the quality measures were quite comparable between the two, although the modeling coefficients differed somewhat.
In summary, i think the new forecasting feature from Tableau will save quite a bit of time and effort for people who are not interested in building sophisticated time series models, but want to get by with some quick, but reliable forecasts. The speed with which one can generate forecasts, provided the underlying data is available within Tableau, simply trumps the need to do R programming, and using its basic plotting features. For those of us who need deeper statistics, such as correlograms or confidence bands around forecasts, we still need tools like R!
Talk to us to find out what analytics tools are best suited for your needs. Sign up for a free consultation.