Troubleshooting Insights

---

Q: When should I use Insights?

Problem	Solution
Need to monitor signals continuously without manually defining thresholds or labels for supervised models.	- Use Insights to automatically learn normal behavior and detect deviations without manual configuration.
Signals are complex, making traditional rules-based modeling difficult.	- Leverage Insights’ AI-based anomaly detection to handle multivariate dependencies and subtle deviations.
Early detection of unexpected events is critical to prevent downtime or operational issues.	- Deploy Insights to catch anomalies before they escalate, providing timely alerts for preventive action.
Difficulty identifying areas of disruptive behavior is signal data.	- Use Insights to pinpoint regions of unexpected behavior, speeding up investigations.
Limited technical resources or expertise in coding/ML.	- Insights requires no coding or machine learning knowledge, enabling any user to set up automated anomaly detection.
Historical data is sparse or insufficient for learning patterned behavior.	- Consider complementing Insights with traditional rules-based monitoring until sufficient data is available for AI-based learning.

Q: Why am I seeing unexpected anomalies or missing deviations?

Problem

Solution

Insights shows unexpected anomalies or missing deviations.

- Insights learns “normal” behavior from the defined learning range. Excluding expected variations may cause them to be flagged later. - Define learning ranges carefully to include all expected normal behaviors. - Abnormal data in the learning range may be treated as normal if heavily present. Update learning ranges when operations change to avoid false anomalies. - The insight may not have had enough time or data to learn the normal patterns and differentiate them from abnormal. Use incremental learning to extend models and reinforce correct patterns.

Q: How should I interpret anomaly scores?

Problem	Solution
Difficulty interpreting anomaly scores.	- Anomaly scores = normalized reconstruction error across sensors. - Higher score = greater deviation from learned behavior. - Use the Insights Dashboard heatmap to locate unusual time periods and signals. - Add reference signals to provide operational context when interpreting anomalies.

Q: How do I extract anomaly data via API?

Problem	Solution
Extracting anomaly data via API.	- Use the Insights dashboard to visualize anomaly signals with the tree hierarchy. - To extract anomaly scores from the application use the REST API. - Focus on key fields:   - firstpoint — timestamp of anomaly start   - max — maximum anomaly score in the window - Other JSON fields are usually unnecessary. - Retrieve ≤1,000 data points per API call for best performance.