Outliers

Outliers

Outliers Overview

Machine learning Forecasting in Arkieva is based on three years of historical sales volume. This history may include events that do not conform with the general characteristics of the historical data. These event may influence future forecast and may be excluded in the forecast estimation. The outlier module automatically detects and categorizes these abnormalities so that you can adjust the history. 

To determine outliers, the engine uses anomaly detection and correction techniques in machine Learning. These techniques examines the shipment history for statistical characteristics such as trend, sparseness, seasonality etc. when searching for anomalies in the data.

Outliers are categorized into one of four levels of severities: Critical, High, Medium, and Low, depending on the potential impact of the statistical forecast.
Use the forecast preview button to evaluate examine how the forecast will be affected if the outlier correction is applied. The forecast in the preview is determined by Arkieva Intelligent Forecast.

Things to consider: 
  1. Outlier scans are performed at the Item and Location level, but are displayed at the (aggregated) Item level, and allow drilldown to the location level.
  2. All Locations are displayed, even Locations without outliers.
  3. When applying the adjusted values at the Item level, it is based on the ratios of the recommended value first and then sales at the ilc. 

Outlier Severity Calculation:
Score=abs(Qty-Lb)/(Ub-Lb)

Where,
  1. Lb- lower confidence bound
  2. Ub -upper confidence bound
  3. Qty- actual demand quantity

If score > 2.0 then Critical,
0.6<score<2.0 then High,
0.3 <score<0.6 then Medium,
Otherwise, Low

Use the Outliers module to 

  1. Identify adjustments to your sales history that should be applied prior to forecasting.
  2. Evaluate how the forecast will be affected if the outlier correction is applied. 
  3. Publish outlier corrections so the adjusted data can be used in other downstream processes like ML Forecasting. 



What is Arkieva Intelligent Forecast?

Arkieva's AI engine first examines the shipment history for statistical characteristics like trend, sparseness, seasonality etc. and then determines the best set of methods appropriate for the data and its characteristics. Each of the methods are evaluated and the best selected.

Working with Outliers

1. Drilldown Chart Data

To filter the Outlier view, click on a bar in the chart, this will drilldown the informational data. Click on an empty part of the chart to drill back up to the topmost informational data.






2. Forecast Preview 

The Forecast Preview helps evaluate how the forecast will be affected if the outlier correction is applied. Click the Forecast Preview button within a row of the table to view that item's forecast.





3. Adjusted Value

Edit the adjusted value and enter your own adjustment under the Adjusted Value column. 






  1. To clear adjusted values, check the checkbox of the adjusted values respective checkboxes, and select Clear Adjusted Value from the dropdown menu next to the Publish button. 








  1. To apply the recommended value to the Adjusted Value column, check the checkbox of the Recommended values respective checkboxes, and select Apply Recommended Value from the dropdown menu next to the Publish button. 






4. Publish to Sandbox Scenario

Select the rows you want to adjust by checking their respective checkboxes, then Publish to save your adjustments in the Sandbox scenario. 







Open/View in Collaboration

You can navigate directly from an outlier flagged at the Item level to the corresponding Item–Location view in Collaboration using the outlier hyperlink.







You can also launch Collaboration from the Item-Location level by right-clicking and selecting Open in Collaboration --> Open in new tab.

























    • Related Articles

    • Actions

      From the Tool Bar, you can access the Actions drawer, which houses many useful application commands: View Log, Start of Planning Process, Demand Review, Supply Review, and of End of Planning Process. View Log Click the View Log button to view a list ...

    • Forecasting

      Machine Learning (ML) Forecast Overview Organizations often sell hundreds, if not thousands, of finished products to a variety of customers through a network of warehouses, distribution centers, and production centers. Creating a forecast plan that ...