Explainability provides some visibility into the machine's decision making process. A higher score for a variable indicates that the variable was used more frequently in the model's decision making process during the chosen time range.
Ultimately, the decision making process underlying each model relies on multi-dimensional comparisons between all variables and stocks, which is not visualized on this screen.
Although it may seem intuitive to only use very important variables in your future models, this will often lead to lower performance overall.
Low to medium scored variables may be used as a flag for a good buy/sell decision, but are used less frequently than the higher scoring variables.
Very low or 0 value scores are a good indication that the underlying data of the variable is not predictive in your current model.
Experimentation with different variable sets will help identify the best available data for predicting good results in your model.