Updated at: 2022-12-09 03:49:50
Time-series forecast can predict the future by existing historical data. For example, according to the historical access log data of the service, the access request data in the coming week is predicted, and the operators can analyze the problems such as user retention and tap into the essence of the problem according to the prediction results.
A complete time-series forecast task includes two parts: data preview and model computation: 
• Data Preview: It is to visualize the data and dimensions that users want to analyze;
• Model Computation: It is to provide a variety of algorithms for users to select different algorithms and parameters for model computation as needed.
To create a new Time -Series Forecast, the specific steps are as follows: 
1. Click Machine Learning > New to create New machine learning task, and click Help to view the Brief and Scenario for Show of the time-series detection machine learning task, as follows:

2. Click Time-Series Forecast for parameter configuration, and click Help in the upper right corner to view the brief, usage help, parameter configuration guidance, and algorithm introduction of Time-Series Forecast, as follows: 


3. Configuration data preview: 
The data preview part is to visualize the original data, data statistics and trend graph.
1) Configure data preview: Fill in the data preview configuration information. During configuring data preview, except that only Time can be selected for Bucket, the rest items are exactly the same. For details, please refer to the section One-Dimensional Anomaly Detection ;
2) Click Preview to view the data preview result, as follows: 

The data preview results include trend graph and Raw Data list:
Field Name
Description
Trend The trend of the selected data: 
► Hover over the trend graph to prompt message, where the first line is Bucket field information, and the second line is Metric field information;
► The time range of anomaly trend graph can be zoomed in/out: 
• Hover over the trend graph, slide the mouse wheel to zoom in/out time range;
• Drag the scaler below the graph to zoom in/out.
Raw Data The first column of the Raw Data is Bucket field information, and the second column is Metric field information
4. Configure model computation: 
For time-series forecast, three algorithms are currently provided: Simple Exponential Smoothing, Double Exponential Smoothing, and Triple Exponential Smoothing. For some algorithm parameters, AnyRobot can automatically adapt to the parameters. To add manually, you can tick Set Manually for algorithm parameter setting.
• Simple Exponential Smoothing: It is a special weighted average method, suitable for the data of horizontal curves.
Field Name
Description
Sliding Window For the size of Sliding Window, the smaller the value, the better the data fit.
Forecast Cycle The set time interval (day). For example: The set time interval is 2d, if the forecast period is 2, it means that 4 days of data will be forecast.
Alpha The parameter of importance decay speed, between 0 and 1. If the data fluctuation is little, set a smaller value; if the fluctuation is large, set a larger value.
Click Compute to view the Time-Series Forecast result, as follows:

Time-Series Forecast model computation result description:
Field Name
Description
Forecast Result The line chart of forecast result, in which there are 3 kinds of line segment to show the forecast result: 
• Original Value: The black line indicates the original value of Metric field;
• Fitted Value: The blue line indicates the value fitted by the algorithm;
• Forecasted Value: The pink line indicates the forecasted result value.
Hover over the trend graph to prompt message, showing the original value, fitted value, and forecasted value at the current time. The time range of anomaly trend graph can be zoomed in/out: 
• Hover over the trend graph, slide the mouse wheel to zoom in/out time range;
• Drag the scaler below the graph to zoom in/out.
Goodness of Fit The goodness of data fit, and the value closer to 1 means high goodness.
Root-Mean-Square Error The goodness of forecast result, and the smaller value means the better model.
• Double Exponential Smoothing: The smoothing is performed again based on the result of Simple Exponential Smoothing, thus preserving the trend information of data. It is suitable for slope-type time-series data and can predict the time series with trend.
Field Name
Description
Sliding Window For the size of Sliding Window, the smaller the value, the better the data fit.
Forecast Cycle The set time interval (day). For example: The set time interval is 2d, if the forecast period is 2, it means that 4 days of data will be forecast.
Alpha The parameter of importance decay speed, between 0 and 0, 1. If the data fluctuation is little, set a smaller value; if the fluctuation is large, set a larger value.
Beta The impact parameter of data trend, between 0 and 1. If the data trend is of long term, set a smaller value; if the data trend is of short term, set a larger value.
Click Compute to view the Time-Series Forecast result, as follows:

• Triple Exponential Smoothing: Seasonal information is retained on the basis of Double Exponential Smoothing, which can forecast time series with seasonality. It is suitable for trend-type seasonal time series data, and is mostly applied to parabolic data.
Field Name
Description
Sliding Window For the size of Sliding Window, the smaller the value, the better the data fit.
Forecast Cycle The set time interval (day). For example: The set time interval is 2d, if the forecast period is 2, it means that 4 days of data will be forecast.
Computation Method • Accumulation model: It is suitable for time series with linear trend and seasonal effect, but not changing with time;
• Multiplication model: It is suitable for time series with linear trend and seasonal effect, but changing with the magnitude of the series
Seasonal Cycle It is mainly applicable to seasonal data, with value between (0,1/2 of Sliding Window].
Alpha The parameter of importance decay speed, between 0 and 1. If the data fluctuation is little, set a smaller value; if the fluctuation is large, set a larger value.
Beta The impact parameter of data trend, between 0 and 1. If the data trend is of long term, set a smaller value; if the data trend is of short term, set a larger value.
Gamma The seasonal impact parameter, between 0 and 1. If the data trend is of long term, set a smaller value; if the data trend is of short term, set a larger value.
Click Compute to view the Time-Series Forecast result, as follows:

5. After completing the above configuration, click Save to fill in the machine learning task Name, and click OK to complete the machine learning task creation.