I develop algorithm for anomaly detection problem, that use also the HTM engine.
I encounter in problem for estimating the accuracy of the algorithm .
The algorithm is working well but the algorithm recognize the anomaly with phase ,
the phase is changing .
y_actual = [0,0,0,0,1,0,1,0…,1,0,0,0,…,0,1,0,0,0]
y_predict = [0,0,0,0,0,0,0,1,…,0,0,0,1,…0,0,0,0,1]
The algorithm recognize/predict anomaly in the area of the actual anomaly but not in the specific position (with different phase) , for my purpose it’s fine .
When I trying to compare the results by position I received poor results in the standard estimation method (TPR=72%/TNR=98%) , when I am looking on the results manually I see the algorithm is more accurate … (when there no anomaly the algorithm return TNR=100,TPR=100)
May someone can recommended on methodology for tuning the estimating accuracy.