I am using an LSTM autoencoder to detect anomalies in a time series. I use the reconstruction error to check if a point is anomalous or not. However, my question is: can I use the precision_recall_curve from Python and give as arguments the truth values with 0 and 1 and the MSE that my model provides? Will I get precision/recall and thresholds related to the MSE in this case? I am confused since I don’t have a regular classifier that outputs probabilities.
