Process for Developing Robust AI Models & Choosing AI Feature Inputs

AI Machine Learning
1

Does anyone have a good process for developing AI Models that Work well Out of Sample in P123?

What I have done so far is run models for each type of feature, example Growth or Quality using the P123 factors available. By doing this I am trying to determine which Features/Feature combinations have strong predictability.

Once I have run a model, example I just created a model using only Growth features I run the predictor and check the importance sorted by co-efficient.

Is it useful to then go through the list and reduce the number of features, and remove the obviously more correlated features, than throw them back into the mix to create a model with only the better features for each "features type". By feature type I am growth, or value, or quality etc.

Or should I choose select everything and throw it into the mix?

I don't have any of my own formulas or factors, is the list default to P123 sufficient?

2

If you're just starting out with machine learning, you don't need to worry too much about feature selection. You need to be more concerned about what universe you are training in and what algorithms you are using.

Honestly, I still can't find a method to select feature to beat the simplest way you can image.

1 Like