Many organizations today are relying on finance and sales business users to provide business forecast for upcoming seasons. You might have historical data for past few quarters or even past few years. Forecast is part of predictive analytics. Predictive analytics is used to make estimations about unknown future events, based on many techniques from machine learning, modeling, and artificial intelligence. The patterns found in historical and transaction data can be used to identify risk and opportunity. Predictive analytics helps business today to make decisions, become proactive, and plan ahead.

Sample data

In the following sample data, I have daily sales records collected for the past four years, with a total 9,500 records. In data preparation, you can examine and understand the data values.

Method 1: Visualization

You can create the monthly sales statistics by adding month and sales to the visualization. As you can see, the sales results have fluctuated in the past four years.

With a right-click, you can add the forecast function to forecast the sales results based the historical sales data.

The forecasting result for sales is generated for the next three months in the shadowed area with a blue shade indicating the possible value based on 95% prediction interval. You can adjust the results by choosing different forecasting algorithms.

In Oracle Analytics, we have the following time series forecast algorithms available out of the box:

• Auto-regressive integrated moving average (ARIMA): This default setting assumes that past data is a reliable base to explain and project the future.

• Seasonal ARIMA: If you find a regular pattern of changes that repeats over time periods, choose Seasonal ARIMA.
• Exponential triple smoothing (ETS): This option is often used in analysis of time series data.

You can also change various options, such as number of periods and prediction intervals applicable to the business results.

Method 2: Data flows

We can also perform a sales forecast with data flows in Oracle Analytics. Data flows are series of steps that help us to organize and integrate data. For example, we can filter the data and select the columns that are required. For advanced users, we can also transform columns with series of data functions.

Using the same sample data, you can create data flows with a step to perform time series forecasts. More parameters are available to you to define the forecast results. In the example, the data is at the daily transactions level. So, I set the period to generate sales forecasts for the next 90 days. Again, you can choose a forecast algorithm of seasonal ARIMA, ARIMA, or ETS.

Last, add another step to save the result. Running the data flow, the prediction result is saved as a new data set, which you can preview in the visualization project.

As demonstrated, you can view a more detailed sales forecast generated from data flows and create a monthly sales forecast from it.

Conclusion

Forecast with machine learning is not complex with Oracle Analytics. If you’re ready with gathered data, feel free to try these approaches out to perform forecasts for your business.