3 years ago
Better anomaly detection accuracy
We have improved the anomaly detection accuracy (and reducing the number of false-positive alerts) by automatically detecting seasonality (e.g. “spend patterns”) in your cloud usage and cost. In the attached examples, the suspected anomaly is a significant increase compared to previous days. However, our new forecasting model detects the cost is within the forecasted range, and the variation is due to historical trends and seasonal effects.
Consider the following (real-life!) examples:
- the orange dot is the current daily cost which is the anomaly suspect
- the black dots represent historical data
- the blue line is our new forecasting model
- the blue region/band is the forecast range, i.e. upper/lower range of our prediction