Overview

Sifflet machine learning (ML) models let you add your own input to their training information in the form of a feedback loop.

Description

Qualifications

By providing feedback on alerts, you can improve model accuracy. Currently available qualifications are:

• False Positive / Expected: A data point was falsely detected as an anomaly. This can be used to flag points just outside the confidence band or to tell the model that the trend has changed, for example when entering a busy period, and avoid anomalies.
• False Negative: There is an anomaly in data, but the model has not raised it.
• Reviewed: A neutral qualification indicating that the error has been analyzed and there is no further action to take.
• Fixed: The anomaly has been fixed. Sifflet can automatically qualify data points as fixed if they were previously anomalies and their value has changed and their updated values fall within the expected confidence bounds.
• No Action Needed/Known Error: Use this qualification to flag known issues that won't be fixed or don't need to be fixed. The model ignores this data point.

📘

Data point qualification influence on the model

Some influence

• When qualifying a data point as False Positive / Expected or False Negative, Sifflet updates the predictions to attempt to detect/not detect similar points as anomalies
• When qualifying a data point as No Action Needed/Known Error, the model ignores it during the prediction process.

No influence

• When qualifying a data point as Reviewed or Fixed, the model acts as normal and does not change its behavior based on the qualification.

Usage

To qualify a data point, click it. A qualification modal opens.

Examples

Example 1: Known Error

In the case below, Sifflet detected big drops in September. After investigation, your team identified the root cause but won't fix it for now. Since it won't be fixed any time soon, and to avoid impacting future predictions, you can set the qualification to "No Action Needed/Known Error".

Example 2: Expected Business Shift

In the case below, Sifflet raised an alert. After investigation, it appears that the alert was inaccurate because the increase and change in trend is an expected business shift.

To improve future predictions and avoid anomaly noise, qualify the data point as "False Positive / Expected". As the example shows, the model quickly tries to adapt to higher expected values.

Example 3: Qualify a Data Point as False Negative

If Sifflet has failed to detect an anomaly, you can qualify a passing data point as False Negative. This impacts the anomaly detection algorithm for future data points to avoid considering similar data points as passing.

By clicking any data point in the graph you can display the qualification window and select False Negative from the list.

Alternatively, and particularly if you need to do this for several data points, you can also navigate to the list of data points below the graph and select the data point to qualify from the list directly.

In this case, the qualification window shows the list of selected data points.