I am working on a classification algorithm to recognize a sound anomaly. I have two audio classification systems placed at different locations. I would like to interconnect these systems and base my results based upon these simultaneous systems. I am giving an "anomaly score" to the classification algorithm results. If both of my systems agree, I can confidently say that an anomaly is detected, however if one system detects an anomaly and the other doesn't how could I deduce the correct result? I can empirically work out an algorithm however, I would like to know if there is a systematic approach to this problem. Does anybody know what this type of problem is called is signal processing? Hence I could further reach on this matter. Also what kind of evaluation metric could be used to describe the system?

  • $\begingroup$ That's a decision problem. You're not giving enough information on what you're detecting, and how you're detecting it, so in-depth advise is impossible... $\endgroup$ Sep 27 at 11:26
  • $\begingroup$ but think about this: what is the thing you want to minimize or maximize with your combined decision, under what boundary condition? Often you want to do something like "as long as I'm not missing more than 1 in 1000 anomalies, I want to reduce the false alarm rate as far as possible", or "I can deal with 5% false alarms, and I want to miss as few anomalies under that condition". Or, you always want to have "I want my decision whether there is an anomaly or not to have the highest likelihood to be correct, given what I've observed". Note that these three are very different optimality terms! $\endgroup$ Sep 27 at 13:06

Your Answer

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

Browse other questions tagged or ask your own question.