Understanding Forecasting
Video tutorial
Forecasting is a method of extrapolating or predicting data based on time. BIRT Analytics forecasting uses the Holt-Winters method, iteratively applying a formula to produce a time series and a forecast. This formula uses a weighted average of data prior to time t to provide a result for time t.
This method consists of three components: the level, trend, and seasonal component.
For example, to forecast the number of orders to be received during the next 12 months, you would perform the following tasks:
*
Select the data you wish to observe, the y-axis of the time series.
*
Select the time interval for the series, the x-axis of the time series.
*
Execute pre-analysis, if necessary.
*
Set model parameters. These may include:
*
Number of projections to make
*
Whether the time series has seasonality
*
Seasonal periodicity
*
Whether or not to replace outliers
*
Level, trend, and season smoothing parameters
*
Execute forecast.
More about outliers
Outliers are observations that appear to deviate markedly from other members of the sample in which they occur. When running pre-analysis, values that are more than two standard deviations away from the sample mean are considered outliers. Outliers are replaced by the sample mean. Generally, outliers should be replaced so that they do not bias any projections.
The value of the previous observation replaces any null values encountered in the sample. If the null value encountered is the first observation of the sample, the value of the nearest non-null observation replaces the null value.
How to create and execute a forecast
To get a forecast of the number of orders to be received in the next 12 months, select Parameters and complete the following procedure:
1  
From My Data, drag the Order table and drop it in Measure. Leave the default operation as Count.
2  
Drag the Month column and drop it in Column, in the Dimension area.
3  
To find out if your data contains a seasonal element, select Has seasonality and set Seasonal periodicity to 12. With monthly data, there are 12 observations in each cycle until the seasonal pattern is repeated.
4  
Set other values, as required:
*
Filter: Constrain the data by the specified selection.
*
Number of projections: The number of observations to make into the future.
*
Replace outliers: To eliminate outliers in the original data.
*
Format: To specify the number of decimal places to use when displaying the forecasted data.
*
Autocalculate: To calculate the smoothing parameters for the model automatically.
*
Level, Trend and Season smoothing: If Autocalculate is not selected, you can set the smoothing parameters for the model manually. Smoothing parameters are values from 0.1 to 0.9. The smaller the smoothing parameter is, the less weight given to the most recent observations in the time series will be, making the series smoother.
5  
Execute pre-analysis:
1  
To analyze the time series before making a prediction, choose Pre-Analysis.
2  
On Pre-Analysis, analyze the results and decide:
*
Whether to replace any outliers in the time series. The number of outliers appears next to Number of outliers. The coordinates of any outliers appears on the graph with a red line.
*
How much confidence you can place in the projections. Evidence for non-randomness is a 5-star assessment. An assessment of three or more stars is considered statistically significant. The less random the results are, the more confidence you can have in the reliability of any projections made. Keep in mind, however, that any randomness may be attributed to seasonal patterns in the time series.
*
Whether to account for seasonality in the forecast and, if so, the seasonal periodicity. Evidence for seasonality is a five-star assessment. An assessment with three or more stars is considered statistically significant. If seasonality is indicated, you select Has seasonality when making a forecast, using the same seasonal periodicity that you used to run the pre-analysis.
6  
To run the forecast, choose Calculate.
How to use forecasting results
After calculating the results, you can analyze and save the forecast.
1  
On Forecast, analyze the graphical results:
*
A yellow line represents the forecast with a blue line representing the original data. Hover the cursor over a chart co-ordinate to display its value. All projections are labeled numerically on the x-axis, starting from 1.
*
If outliers were replaced, then the original data shown displays the value used to replace the outlier.
*
You can also zoom in to any part of the chart to get a more detailed perspective. Select the area of the graph on which to zoom. Zoom out by right-clicking in the chart and selecting Zoom Out Chart. Choose Reset Chart to show all data points.
*
To make comparisons between different time periods, overlay different parts of the chart. Choose Switch to pin mode and select an area to float, then drag the floating area over another area of the chart to make the comparison. To return to the original state, choose Switch to Zoom Mode.
*
The smoothing parameters used for the model, including those selected automatically, appear at the top of the tab.
*
To export the forecast to a rich text format (RTF) document, choose the printer icon. The generated document contains the chart exactly as shown in the Forecast tab, along with the data in tabular format.
2  
On Table, review the forecasted data and the original data in tabular format. The forecasted values are labeled numerically, starting from 1.
3  
To save the forecasting analysis, choose Save. On Save, you can choose to share the analysis with other BIRT Analytics users.
Video tutorial
Predicting seasonal trends in your data
Additional Links: Forums | Blogs | Support | Give Us Feedback | More...

Copyright Actuate Corporation 2013 BIRT Analytics 4.2