I used STL decomposition on the power consumption data from an air-conditioner over a period of 10 weeks. I would expect this data to be periodic over a week. What I am able to observe from the data is that the curve has huge residual values when compared to seasonality or trend.
Does this mean that the data that I have does not accurately represent the weekly cycle of air-conditioner usage? Or is this model still good enough to be used for anomaly detection? Also, the trend seems to have a periodicity. What does this signify?
Case Closed.
It was indeed the case that the data does not accurately represent the weekly cycles of AC usage. The AC was being used on whims.
