Tasks: Anomly Detection TimeSeries Data

[mai1x9]

• Anomly detection within timeseries data. We will be having quite a metrics data like lets say 10 columns, all are metrics data, 11th column is timestamp. These metrics are generated at certian frequency
• Now I need to understand the anomlies within the metrics data for all 10 cols based on past history and identfy the patterns. Eg: cpu utilization is one column data. Now lets say u got time seriews of cpu utlization over past 1 week, now if u see sudden incremease in value then its anomly.
• And we also have text data that gets generated at certian freq. Any new text pattern identified it must be notifed as anomly. So we dont have metrcis data rather text data generated by the sensors. Eg: the best example would be like, think of ipad tracking time feature, it gives time spend on each application, in our case we have entire history of applications used with timestamps and time spend on each application. Now if completely new application is seen then its anomly. Eg: 2 and at same time an application can have multiple tabs. lets say browser tabs. All are under chrome application but i can visait any website. lets say i visit a website which i shouldnt or i never visited that website or similar websites from my past hsitory so its an anomly

---

Libraries to be used

• https://github.com/chaos-genius/chaos_genius for reference and understanding
• https://github.com/facebookresearch/Kats/ facebooks time series model.
• Alibi detection - anomly detection: https://github.com/SeldonIO/alibi-detect
• darts: https://github.com/unit8co/darts
• Forecasting: https://github.com/Nixtla/statsforecast
• Merlion https://github.com/salesforce/Merlion for reference
• Orion: https://github.com/sintel-dev/Orion timestamp and simple metric value.
• Auto encoder based anomly detection: https://github.com/datablogger-ml/Anomaly-detection-with-Keras

[mai1x9]

For reference: https://github.com/rob-med/awesome-TS-anomaly-detection