Viewing Project Insights
When you open an active project, you see the Project Summary view. To access project insights, select Insights from the View menu. You can take these actions from the Insights page:
- Click Monitoring details to see the monitoring workflow and access logs for each node in the workflow.
- Click Dashboard to see how your model is performing.
- Click Scoring Insights to see details about scoring.
Dashboard Metrics
Click either Performance
metrics or Fairness metrics to select which
insights you want to view. The table lists the performance or fairness metrics that
you
selected to track for the project. The trend column represents the change in the metric
value since the previous run. If the value is increasing, the model is likely performing
well. A decreasing value might indicate that your model is losing performance or
starting to drift. You can also access a description of each metric by clicking 
.
The following sections are available when you select Performance metrics.
Trend Over Time
The Trend Over Time graph shows you how different metrics perform over a period of time.
- Click Customize to select which metrics you want the graph to display. If only one metric is selected, a threshold reference line is shown.
- Click one of the values for Past number of days to change the number of days that are displayed in the graph.
Receiver Operating Characteristic (ROC) Curve
The Receiver Operating Characteristic (ROC) Curve evaluates how well a model distinguishes between these two outcomes by plotting two metrics across every possible decision threshold:
- The proportion of actual positive cases that the model correctly identified. (Recall)
- The proportion of actual negative cases that the model incorrectly identified as positive. A curve that hugs the top left corner indicates strong performance on both metrics. A curve along the diagonal indicates the model is no better than random chance. The Area Under the Curve (AUC) summarizes this as a single score from 0.5 (random) to 1.0 (perfect).
For example, for projects that use a Likely to Churn recipe, the ROC curve shows the trade-off between identifying churners (recall) and identifying non-churners (specificity). These insights can help you to fine-tune your efforts so that they have optimal impact. The AUC score ranks how effectively the model ranks everyone by their likelihood to churn, with higher scores indicating better prioritization.
Predictor Importance
The Attribute Importance graph shows the attributes that had the most impact, either negative or positive, on the outcome. The predictors that are evaluated come from the recipe that this project is based off of.
Overfit Check
The Overfit Check section shows you whether the model performs consistently across data that it learned from and new data. A difference in accuracy greater than 5% between the two sets might indicate overfitting.These are the possible results:
- Passed means that the model shows no sign of overfitting, suggesting it likely generalizes well to new data.
- Caution means that the model shows some inconsistency. You might need to review it to ensure reliability.
- Failed means that the model appears to be overfitting, suggesting it might not reliably generalize to new data.
Scoring Insights
Click Scoring Insights to view how scoring is performing. The following sections are available:
- The tiles in the Latest Run section give you a quick view of how the last run is performing.
- The Distribution of Scores graph shows you the score threshold and the distribution of scores.
- The Scoring History table shows you
data about previous runs. You can take these actions:
- Click View for a run to see its details.
- Change the time range of the graph by clicking one of the options. All is selected by default.