What is Microsoft Time Series Algorithm

The management team at Adventure Works Cycles wants to predict monthly bicycle sales for the coming year. The company is especially interested in whether the sale of one bike model can be used to predict the sale of another model. By using the Microsoft Time Series algorithm on historical data from the past three years, the company can produce a data mining model that forecasts future bike sales. Additionally, the company can perform cross predictions to see whether the sales trends of individual bike models are related.

Each quarter, the company plans to update the model with recent sales data and update their predictions to model recent trends. To correct for stores that do not accurately or consistently update sales data, they will create a general prediction model, and use that to create predictions for all regions.

In SQL Server 2005 (9.x), the Microsoft Time Series algorithm used a single auto-regressive time series method, named ARTXP. The ARTXP algorithm was optimized for short-term predictions, and therefore, excelled at predicting the next likely value in a series. Beginning in SQL Server 2008, the Microsoft Time Series algorithm added a second algorithm, ARIMA, which was optimized for long-term prediction. For a detailed explanation about the implementation of the ARTXP and ARIMA algorithms

By default, the Microsoft Time Series algorithm uses a mix of the algorithms when it analyzes patterns and making predictions. The algorithm trains two separate models on the same data: one model uses the ARTXP algorithm, and one model uses the ARIMA algorithm. The algorithm then blends the results of the two models to yield the best prediction over a variable number of time slices. Because ARTXP is best for short-term predictions, it is weighted more heavily at the beginning of a series of predictions. However, as the time slices that you are predicting move further into the future, ARIMA is weighted more heavily.

References:

https://learn.microsoft.com/en-us/analysis-services/data-mining/microsoft-time-series-algorithm?view=asallproducts-allversions#example