I've been reading about Matrix Profiles and how they can be used for anomaly detection in time-series. However, I'm a bit confused how they compare/relate to typical Deep Learning approaches. I would imagine Matrix Profile would take less time to train, but I have no evidence for this.

How does Matrix Profile compare to Deep Learning in terms of anomaly detection in multi-variate time-series?

Related to "Are time series motifs and the Matrix profile algorithm a good fit for my problem?" wherein the author asks about using the Matrix Profile for anomaly detection and similarity between two signals.


According to "Time is of the Essence: Machine Learning-based Intrusion Detection in Industrial Time Series Data" by Anton et al. Matrix Profile does indeed require less training time/data and parameter tuning than Deep Learning. Additionally, the Long Short Term Memory (LSTM) they compared to, performed worse than Matrix Profile.

Comparison Methodology

Matrix Profile

Matrix Profile is basically a dimension reduction approach, anomalies are quantified by their distance from existing data. Consequently, a threshold needs to be set. If an ideal threshold is set, the Matrix Profile performs perfectly.


The LSTM is used as a predictor, wherein the difference between the predicted next value and the actual value is used for anomaly detection. Even if an ideal threshold is set for this prediction value, the LSTM does still makes mistakes.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.