Time Series Anomaly Detection Algorithms
Pavel Tiunov

Thanks Pavel for this useful article.

At Expedia we’ve had success applying the STL algorithm to bookings time series. The models are multiplicative rather than additive, but we use a log transform and back-transform to deal with the fact that STL does additive decompositions. Also we handle multiple seasonalities (daily + weekly) using a two-pass approach, where we run STL on the original time series to isolate the daily seasonality and then re-run it on the trend component to isolate the weekly seasonality.

If you are interested in learning more please see


One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.