Applying preprocessing transformations on features
At this moment we are making predictions without optimizing the datasets features.
The first step would be to tags them. some categories have been discussed :
- Constant
- Statuses
- Variable
These tags could be applied automatically or by experience.
Then based on this, some process will be applied on those features (that part have to be defined yet)
- Handling missing values
- Handling possible outliers
- Categorical feature (at this time we drop all no numerical features, but maybe we will have to deal with interesting categorical ones)
- Creating new features from existing ones
- Transforming existing features (grouping, ...)
Edited by deck