I am working on a Forecasting problem. This is my procedure:
a) Standard scaler (supports partial fit)
b) Detect outliers using DBSCAN (does not support partial fit)
c) Forecast with Kalman filter (not sure of this???) or MLP (supports partial fit)
So, I would like to know if this procedure is possible to use "partial fit" (incremental learning)?
Or do all algorithms have to support partial fit?
Can you add more detail on your usecase? Also, I would recommend you to use StateSpace algorithm for forecasting as it supports partial fit. Check the documentation here: https://docs.splunk.com/Documentation/MLApp/4.2.0/User/Algorithms#StateSpaceForecast
I am predicting the number of logins. My dataset has the number of logins by hour (1 month).
I use a) and b) to clean the data (removing or transforming outliers).
With c) I forecast
a) Standard scaler
b) Detect outliers using DBSCAN
c) Forecast with Kalman filter or MLP