KDDCup2021

MSBD5002 Data Mining course project

---

Required packages

• Numpy, Pandas, Scipy
• rrcf for robust random cut forest
• stumpy for matrix profile
• sranodec for spectral residual

---

We ensemble several methods for anomaly detection, including statistical methods (statistic_func.py), fourier transformation （fourier_transform.py）, matrix profile (matrix_profile.py), spectural residual (spectral_residual.py) and robust random cut forest (rrcf.py). Details of each method cound be found in each of the python file.

To run the program, use the following command

bash ./run.sh

It may take a few days due to the significant resources required to compute the matrix profile, although we use a 6-core GPU.

The ensembled results are saved in ensemble_results folder. We take the index with the largest confidence score to be the final detected anomaly location. The final result is in submission.csv

References

• Time Series Anomaly Detection
• Multi-dataset Time Series Anomaly Detection