FluxEV

The code is for our paper "FluxEV: A Fast and Effective Unsupervised Framework for Time-Series Anomaly Detection" and this paper has been accepted by WSDM 2021. [中文解读]

😉 About Name: "Flux" means the Fluctuation, "EV" denotes the Extreme Value.

💫 Good News: FluxEV has been integrated into OATS, a convenient Python library providing various approaches to time series anomaly detection. Thanks for their awesome work!

Requirements

• numpy
• numba
• scipy
• pandas
• scikit-learn
• matplotlib (plot)
• more-itertools (plot)

Datasets

1. KPI.

• Competition Website Link: http://iops.ai/dataset_detail/?id=10 (⚠️ Seems like the official competition website is down!)
• Tsinghua Netman GitHub Link: https://github.com/NetManAIOps/KPI-Anomaly-Detection/tree/master/Finals_dataset

2. Yahoo.

• Official Website Link: https://webscope.sandbox.yahoo.com/catalog.php?datatype=s&did=70

Instructions

preprocessing.py:

• Fill the missing points for KPI dataset.

spot_pipe.py:

• SPOT function is modified to be a part of FluxEV for streaming detection;
• MOM(Method of Moments) is added as one of parameter estimation methods;
• For the original code, please refer to SPOT (Streaming Peaks-Over-Threshold)

eval_methods.py:

• The adjustment strategy is consistent with AIOps Challenge, KPI Anomaly Detection Competition.
• The original evaluation script is available at iops.

main.py:

• Implement streaming detection of FluxEV.

Run

python main.py --dataset=KPI

python main.py --dataset=Yahoo

Citation

@inproceedings{li2021fluxev,
  title={FluxEV: A Fast and Effective Unsupervised Framework for Time-Series Anomaly Detection},
  author={Li, Jia and Di, Shimin and Shen, Yanyan and Chen, Lei},
  booktitle={Proceedings of the 14th ACM International Conference on Web Search and Data Mining},
  pages={824--832},
  year={2021}
}