What is the most appropriate machine learning model to detect abrupt changepoints in time-series data?

Question

I have a dataframe with a column for time in which anomalies occur. The anomalies are of a sudden changepoint in the time column.

The changepoint is the blue point in the graph.

My goal is to identify these points as anomalies and mark them.

I've tried searching for anomaly detection ML models for this problem such as:

• Low pass filter
• SVM
• AR
• ARIMA
• BOCD

None of these ML models succeed at identifying these anomalous points but maybe i missed something.

Any help will be much appreciated.

Answer 1

With an anomaly detection algorithm like OneClassSVM, GaussianMixtureModel, IsolationForest etc your training setup and features will be the deciding factor in success or not.

The kind of anomaly shown there will be easily detected if you transform your data to differences or rate-of-change between consecutive time-windows.