SHAP Value with one-hot encoding by LightGBM

[hugolam1]

I am struggling with 2 main focuses in my modelling work:

1. Should I use one-hot encoding in LightGBM to show all the category's importance, would it cause multicollinearity?

2. How can I show the SHAP value importance if I used one-hot encoding in my dataset and how do I know if the categorical feature has a positive/negative impact?

I saw some of the posts wanting to group back the categorical feature after the encoding, but I would like to focus on all categories in the categorical features.

For example, the variable: "industry" - the categories after one-hot encoding are: industry_finacnce, industry_tech and industry_manufacturing.

My goal is to show these 3 categories in the shap value importance plot.

Does anyone have any idea about these? Appreciate your help in advance!!