Machine learning widget types
You can use the machine learning widgets in Analytics to analyze the results obtained from machine learning. To add machine learning widgets into a dashboard, go to Edit > Add AI widget, and then select the required widget.
ⓘ Note: You can add these machine learning widgets to any dashboard. However, the widgets display information only if the following conditions are true:
- Machine learning is enabled.
- Training has completed for the machine learning models. For more information on training models, see the Machine Learning topic.
Influencers & Predictions
This widget displays the predictions and explanations for the predictions (influencers) produced by the machine learning algorithm. The table reports a series of information:
Case ID Running or completed.
Last Perfomed Activity The last activity that was performed in the case.
Elapsed time The time elapsed from the beginning of the case. This column is available only for running cases.
Expected total time The lead time predicted by the machine learning model. This column is only available for running cases.
Expected vs Target The difference between the Expected Total Time and the AVG (Target) time. AVG time is the average duration that was observed for the completed processes, which is available from the Overview widget. The negative difference (indicated by green color) indicates that the case is predicted to finish earlier than the target lead time (the case is on time). The positive difference (indicated by red color) indicates that the case is predicted to finish later than the target lead time (the case is late). This column is available only for running cases.
The same table with different labels is available also for costs.
The bar chart illustrates the influencers of the predictions related to the cases that are selected by the current set of filters. This information is also displayed in the table.
Each bar is an influencer and the x axis represents how much it contributes to increase (if positive) or decrease (if negative) the prediction. The color indicates the frequency of the influencer (the darker the blue the more cases it is related to).
In particular, the value that is displayed in the x axis is the average contribution of the influencer, considering all the cases that it affects.
The bar chart can be sorted by contribution of the influencers or by frequency, in descending or ascending order.
When you select a case in the table, the information related to the case and the influencers that are related to the prediction of that case are displayed in the bar chart.
This is a detailed analysis of the selected case. Information such as influencers and their directions might vary depending on the case selected.
The effect of an influencer on the selected case or a smaller set of cases might be different from its effect on a larger set of cases. This change is because of the interaction with certain variables that are applicable to the selected case. The change in direction in the bar chart indicates this change in effect.
When you select an influencer and then click on the filter icon , all cases affected by that influencer is selected. All the other influencers that are displayed reflect
this new subset of cases.
The configuration of the widget presents several options:
Title
Measures
The available predicted measures (for example lead time, cost).
Data type
Whether you want to visualize and analyze running or completed cases.
Type of filter
How to select the bars to display in the chart, the options are:
- Max number of rows: maximum number of bars displayed in the chart.
- Minimum absolute value: set together with the relevant Units (for example, €, and days) to display only those bars that after the aggregations have a higher absolute value than the provided one.
- % of Average measure: It works as the previous option, but instead of freely setting a minimum value you can set it as the percentage of the average value for the predicted measure (for example average completed lead time = 10 days, Minimum absolute value = 1 day and % of Average measure = 10 will yield the same results).
Max number of influencers
Max number of cases
Prediction summary
It displays the number of running cases, together with AVG Completed Time and AVG Completed Cost. These values are used as references to predict the deviations in AVG Expected Time and AVG Expected Cost.
It also displays the last time at which the machine learning model was trained.
Prediction avg
It is a pie chart that represents the number of running cases based on whether their predicted lead time and cost are over or below the reference targets (AVG Completed Time and AVG Completed Cost).
Displays also other several additional information:
AVG Expected The average Lead Time or Cost predicted by the machine learning model.
Target
The reference value that is used to compare with the AVG Expected values and predict the deviation in lead time and cost of running cases.
Below - AVG Expected vs Target
A comparison between the values below the target and the expected average.
Over - AVG Expected vs Target
A comparison between the values over the target and the expected average.