Problem of prediction


hey guys, I have some understandable keys and I need some of help.
what I know for detect anomaly:

  • we training our model with normal data that mean create a model param with the normal data
  • then for the tests, we compare the current data with the pretected data. If they are identical, the current data are note anomalies, if not so it 's an anomaly. This data prediction is created by a mode paraml and this model is created with the normal data.
    but if that’s true, why my plot is like this, the anomaly data is also protected



My guess is that TM is bursting due to unseen pattern (and generating high anomaly score). Thus TM predicts some value but you also see anomaly flagged in the bottom graph.


yes, I know that my code detects some anomalys, but I think it’s a bad result because there is a problem of prediction.
so I think the broblem is in my param model