How Forecasting Works in Tableau: Creating a Forecast

Forecasting is about foreseeing the future value in each model. Many mathematical models are used for forecasting. At the same time, Tableau is one of the commandings, and precise data visualization tools mainly focused on generating for the Business Intelligence Industry. Data forecasting is vital to predicting, and it acts based on the business intelligence generated. Tableau can be used with many data sources such as text, No SQL databases, Excel, SQL databases, etc.

Learn online data science courses from the World’s top Universities. Earn Executive PG Programs, Advanced Certificate Programs, or Masters Programs to fast-track your career.

Exponential smoothing is the technique used for forecasting in Tableau. Forecast algorithms try to establish a regular pattern which can be adopted for future continuity. You naturally add a forecast into the view, which contains a data field consisting of one measure. However, Tableau creates a forecast to view if data is not present, which consists of dimensions with integer values and at least one measure.

upGrad’s Exclusive Data Science Webinar for you –

Transformation & Opportunities in Analytics & Insights

Forecast Algorithms

Real-world data generated requires simpler algorithms to yield better results and are easy to be used, which is available with all forecast algorithms in Tableau. To generate a high-quality forecast, both the pattern generated by the model and real-world data forecastingshould match reasonably. Here quality metrics are used to measure this match. If the quality is low, the quality band confidence band and quality score are not considered because of inaccuracy in the testing estimates. 

Forecasting in Tableau, Tableau repeatedly chooses the top, up to eight models; the highest quality forecast one is chosen first. Before Tableau assesses forecast quality, the smoothing parameters of each model are optimized.

The optimization method is common and used in each model, making it impossible to choose locally optimal smoothing parameters that are not globally optimal. However, primary value parameters are chosen based on best practices and are not optimized. So, there is a chance that initial value parameters are less optimal. 

When there is a limitation of data for visualization, Tableau automatically attempts to forecast with better time-based granularity and then sums the forecast to the granularity of the visualization. Tableau delivers forecast groups which can be calculated from a closed-form equation. Every model with a multiplicative component or with combined data forecasts contains simulated bands, in which every model is based on the closed-form equation. 

Exponential Smoothing and Trend

The forecast model in Tableau uses exponential smoothing. It is a process of analysing data to identify patterns and generate forecasts. While forecasting different, exponential smoothing models are used. Exponential smoothing models are used to forecast future values in a consistent time series of values using biased mean values of past values in the known series. One modest model is Simple Exponential Smoothing; here, computation is based on the smoothed value from a weighted mean value and the last actual value. This method is exponential because more recent values are given more weight, and each value is based on every preceding actual value. 

Seasonal components are operative when the measured value showcases trend or seasonality over the time period on the operative value. The trend is a property of the data to increase or decrease in the given time period. Seasonality is a reiterating, expected variation in value search, such as a fluctuation of rainfall or occurrence of flash floods.

In general, better data forecasting can be yielded when lots of data is available over the time period. Having sufficient data is most important while modelling seasonality due to complexity in the model, and greater precision can be achieved only with more proof of data available. Due to limitations, a forecast cannot be generated from 2 or more data values. 

Seasonality

To estimate the data forecast in Tableau trials for a seasonal cycle typically with the length for the time aggregation among the time series. If aggregated by months, Tableau will check for a 12-month cycle; if aggregated by quarters, the four-quarter cycle is searched, and if aggregated by days, weekly seasonality is searched. If you consider a six-month cycle in a monthly time series, Tableau will search for a year pattern containing two like sub-patterns. If a seven-month cycle in monthly time series is considered, Tableau will search for no cycle at all as they are uncommon. 

Since every selection is selected with automatic a Tableau derives possible season lengths from the given data, the default “Automatic” model type in the Forecast Options Dialog Model Type menu is not changed. While Selecting “Automatic without seasonality” aids the performance by predicting all season length is probing and assessing seasonal models.

Model Types

In Forecasting in Tableau, when you choose the Forecast Options dialogue box, you are provided with the model type used for your data. Here default Automatic setting is provided, and this is optimal for the primary view of the data. Under Custom, you need to specify the trend and season characteristics, among None, Additive, or Multiplicative types.

Source

In an Additive model, each contribution of model components is added, whereas, in a multiplicative model, a minimal component contribution is multiplied. Multiplicative models tend to improve forecast quality from the value of the data significantly, and the trend or seasonality is based on the magnitude of the data. As multiplication is involved in the multiplicative model, none of the values needs to be zero or tends to zero. 

Forecasting involving dates

 When considering time while forecasting, Tableau allows only one main date and Part dates.  But in the case of Part dates, they should belong to the temporal granule. 

There are three formats of dates, but only two of the following are valid in Tableau.

• Truncated dates: Tableau uses a unit of time dimension known as the granularity of dates. Truncated dates are particular dates in a temporal granular. They are generally continuous, and the green color denotes them in the display. These are valid date formats used in forecasting.
• Date parts: They are the discrete version of the date field.  It refers to a particular part of a given temporal granule. These are not continuous like Truncated dates. They are represented in the blue background by different discrete fields. These can be used in forecasting by Tableau. But it requires a minimum of a Year date part. Less than year date part are not considered for forecasting.
• Exact dates: As the name indicates, they are precise and particular points in past data. They have minute details, like March 31, 2022, at 11:23:23.5. They are not valid in forecasting.

Our learners also read: Free Python Course with Certification

Granularity and Trimming

To create a forecast of the data, the time dimension selection is essential to see and predict in the given time reference. Tableau dates provide a range of time units, including day, month, quarter, and year. Granularity is the unit chosen for the date. 

Usually, data will not align precisely with the unit of granularity. For example, if you set the data in the quarter’s value, actual data may stop before the quarter. Further creating a problem while predicting the model, as a full quarter, is considered by the model, which is typically a lower value than an entire quarter.

If you allow data forecasting models to choose data, then the subsequent forecast will be imprecise. Trimming the data is one of the solutions. Trailing periods that derail the forecast are neglected by using the Ignore Last option provided in the Data Forecast Options dialogue box. The default option is to trim/complete one period.

Also Read: 16 Top Data Science Projects in Python You Must Know About

Sufficient Data

Data is the core foundation. Tableau requires lots of data to make precise and correct forecasts. Trend and seasonality are two crucial concepts used in forecasting. For trend, Tableau needs a minimum of five data points. And for seasonality, the data points should be sufficient for a minimum of two seasons, that is for 24 months or 12 months and 5 periods. For example, For a model of a four-quarter seasonal cycle, a minimum of nine data points are necessary. Similarly, a minimum of 24 data points to evaluate a model of two 12 months cycles.

If there are not sufficient data points for reliable forecasting. Tableau queries the data sources and collects relevant quantities of finer granularity. 

• For example, if you have data of less than 9 years. Tableau will retrieve quarterly data and will produce a quarterly estimate. But later it will sum up and display forecasts on a yearly basis, as per the data.
• While quarterly forecast, if the data points are lower than 9 quarters. Automatically, tableau will query the data source and retrieve monthly data and monthly forecasts. And represent an aggregated quarterly estimate display.  The process goes on depending on the available data.

The Tableau carries on these adjustments in the background without any intervention. The forecast display will not change, only the forecast period will show actual granularity.

Conclusion

Time-based data values have their own rules and implications. Using the right volume of data is essential for better prediction and estimations required for future business adaptabilities. Forecasting in Tableau is so feasible that it can be connected to any software and used to get futuristic and rich data predictions based on the data value.  Forecasting in Tableau automatically tries to predict the trend; after reading this article, you should be clear that usage of “Automatic” does not fit all the data values.

Learning Tableau is essential for individuals aspiring to learn and have a career in Data Science and visualization. Understanding the concepts and using the tool is quite tricky for starters, while industry experts’ learning provides the right start. 

If you are curious to learn about data science, check out IIIT-B & upGrad’s Executive PG Program in Data Science which is created for working professionals and offers 10+ case studies & projects, practical hands-on workshops, mentorship with industry experts, 1-on-1 with industry mentors, 400+ hours of learning and job assistance with top firms.