I am working on a regression task where I am trying to predict future values of a stock/resource. At the moment, my model uses a large set of lags as input features and I want to use feature selection (FS) to select only the important lags needed by the model. However, most FS algorithms seem to be based on classification models so I am wondering whether I can create a 'proxy' classifier which uses the same input data as my regression model but whose outputs are now discretized versions of the regression outputs (i.e. simplest case 0=increase in stock, 1=decrease). Would the selected features from this proxy model serve as 'good' features for the regression model or should I only use FS algorithms designed for regression? I would be most grateful for any suggestions, particularly if they are referenced from previous research papers on the topic.
Classification is one of the two types of regression see the attached and it's references. Best D. Booth
I'd recommend to use a suitable feature selection scorer depending on your problem.
You can find more information and an implementation of the FS methods you are looking for here:
https://scikit-learn.org/stable/modules/feature_selection.html
I think you should perform feature selection algorithms for the regression problem. This link may be useful for you. https://machinelearningmastery.com/feature-selection-for-regression-data/
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression