Time series analysis

Time Series analysis attempts to make forecasts based on a series on time-related input data. It consists of numerous techniques that can be applied to either seasonal or non-seasonal data. For more detailed information on this type of analysis, refer to the Data Mining Services chapter in the Advanced Reporting Help.

The Training Metric Wizard provides access to the following settings which determine the type of analysis to be performed:

• Number of time periods in the seasonal cycle: This value allows for the specification of seasonality inherent to the training data, and may be defined as follows:

• Zero (default) or 1: This indicates that no attempt is made during analysis to find seasonality in the training data. In this case, six types of analysis are performed. This analysis is based on Triple (quadratic) Exponential Smoothing and five different types of trends including none, additive, damped additive, multiplicative, and damped multiplicative. Once the analyses are complete, the model which best fits the training data is created.

• > 1: This indicates that the training data has seasonality, consisting of the specified number of time periods. In this case, ten types of analysis are performed. This analysis is based on five different types of trends including none, additive, damped additive, multiplicative, and damped multiplicative. Each trend is analyzed using two types of seasonality, which includes additive and multiplicative. Once the analyses are complete, the model which best fits the training data is created.