Back to list

Time series anomaly detection


Detection of anomalies in time series of cyclic nature data.

To develop the algorithm our developers have used an algorithm of confidence bands estimation with L2 metrics.

The algorithm was suggested in the article “Confidence bands for time series data” by J. Korpela, K. Puolamaki and A. Gionis.

The mentioned algorithm was modified to track the seasonal factor in time series: for that purpose several variants of data preprocessing were suggested and tested; the cross-validation system for optimum training parameters selection based on the specifics of the client’s system was also realized.

To demonstrate the results data visualization was implemented.

Development time
1 week 1 developer