I would like some pointers about the following problem:

I would like to detect anomalies in a pretty huge collection of railway data. Or create a baseline model for detecting future anomalies. The data I have at my disposal exists out of coordinates and speed at the given coordinate (also the time of measurement). Could this perhaps be approached as a regression problem where there's a clear(?) connection between location and speed of a train. For instance a train suddenly moving at snail's pace on a track that from historical data is considered high velocity could be a potential anomaly. If this could indeed be approached in such a way would something like SVM be an option or should I look into other algorithms?


1 Answer 1


There are a number of directions you can take, and there are a number of questions related to this topic (anamoly-detection-for-transaction-data). One question to answer first is whether the data is seasonal or not.

You already suggested a valid approach. Determine which areas are correlated to low and high speeds and than determine when a speed differs. I think it's best to start with a heuristic approach and than try more complicated methods to try to improve. If you want to use an SVM, a one-class SVM may work well (one-class SVM)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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