I would like to know what is the best approach to detect on-line the occurrences of a new steps or changes in a time series.
I've attached a picture of the time series I'm talking about. I would be able to detect the step indicated by the red arrow.
Each data point is collected every 2 seconds and I need to detect the changes in the time series as quick as possible, i.e. considering at most 5 samples ahead and avoiding if it is possible the outliers.
I'm completely stumped on how I might be able to do this - any ideas?
I have read something on Bayesian Analysis but before studying in depth I would like to know if there are other useful approach in term of detection speed, robustness to outliers and detection accuracy.
Thanks for your help.