Hi,
In regards with the missing Consumption Values, what is the preferred/expected strategy please?
- To interpolate the missing values or
- To consider those entries as Anomalies?
Also, in my naïve experience with Anomaly Detection, the training data is usually represented by a bulk set of positive samples sprinkled with very few labeled negative samples (anomalies) to be used for cross-validation/testing/parameter fine tuning. In this case, ALL training data provided seems to be unlabeled (unless I am missing something fundamental).
Is this meant to be a completely unsupervised learning problem please?
Thank you!