MicrosoftML One Class SVM RRS feed

  • Question

  • I am using the MicrosoftML package to do one class svm as an anomaly detection.

    Let's call "good data" as my 1st class and the "bad data" as my 2nd class. (Think of it as not spam and spam).

    I have way more good data than bad data.

    I built my model with the "good" data using rxOneClassSvm and I evaluated the model using a combinations of good and bad data using rxPredict. 

    The results that I get from the rxPredict is Score and I am not familiar with this. I searched on the web to get a better idea of the Score value. I found that Score is the measure of the distance from the "good" data and based on this I created a factor called predictions to indicate that if the Score >1 then it is "bad" otherwise it is "good".

    As I stated previously I am not familiar with the Score concept. Is my understanding in the previous paragraph on the right track? If not could you provide links to where I can get a better understanding of Score (how is it calculated, what is the difference between the negative and positive score values, Is the score value related to "Score.good" or "Score.bad" ).

    Thank you in advance.

    Thursday, May 18, 2017 7:04 PM