Experiments with anomaly detection and forecasting algorithms based on the ARIMA models

The idea behind this algorithm is to use ARIMA weights as features for the anomaly detection algorithm. Using discrete differences of weight coefficients for different heuristic methods for obtaining function, which characterized the state (anomaly, not anomaly) using a threshold.

Links at PyPi, GitHub.

Numerical experiments for the anomaly detection algorithm

Numenta Anomaly Benchmark

The results are reproducible, but for this, you need to clone the GitHub repository and put the files from the following Folder into it.

Real-world data from power plant equipment

Folder of experiment ❗️❗️❗️The results are not reproducible due to the lack of proprietary data in the repository.

Numerical experiments for forecasting algorithm

Folder of experiment ❗️❗️❗️The results are not reproducible due to the lack of proprietary data in the repository.

Others

Supplementary materials, which are used in the paper