This repository including notebooks with examples about timeseries anomaly detection.
YouTube course https://www.youtube.com/watch?v=92EF4vqaBSE&list=PL7GGfr9mTeYWniRK11xuFsEky07oUQ_tX&index=2
Our dataset is timeseries of food retail:
| ou | datetime | cheques | rto | n_sku | cnt | cashnum |
|---|---|---|---|---|---|---|
| 468 | 2019-11-16 08:00:00 | 34 | 8003 | 137 | 173 | 3 |
| 468 | 2019-11-16 09:00:00 | 40 | 20129 | 283 | 517 | 2 |
- ou - index of shop
- datetime - ISO format of date and hour
- cheques - count of payment
- rto - revenue in rubles
- n_sku - count of lines in bills
- cnt - number of items
- cashnum - number of opened windows while hour
We explore few approaches for anomaly detection in 1-D timeseries such as:
- statistical anomalies based on normal distribution
- forecasting method - detection anomalies as error of forecast
- classification method Isolation Forest
- clusterization method K-means
- K Nearest Neighbors
Also we explore few approaches for anomaly detection in Multi dimensions timeseries based on PyOD library.