Anomaly detection using the Minimum Covariance Determinant (MCD) method

Assume we need to detect anomalies in Gaussian-distributed data or at least with an unimodal, symmetric distribution. First, we fit a minimum covariance determinant linear model. Then, we compute the Mahalanobis distance and consider it as outlier score for each observation.

You can read more details about this methid in this paper:

https://arxiv.org/abs/1709.07045

You can test the code here:

https://github.com/multiopti/MYWAI/blob/main/tsclass_mcd.ipynb

Feel free to leave your comments here below, I would be happy to answer.

At?MYWAI?we promote agile, explainable, reliable and affordable ML at the edge.