如何在数据中找到特征重要性？

Question

I want to know what column will correlated and impact to no_of_purchased but i have both numeric (eg total_item) and non-numeric data (eg shop_type)

**table of data (columns name) **

1. shop_id
2. shop_type (eg franchise,..)
3. total_item
4. is_in_business_district
5. is_creditcard_payment
6. total_staff_in_shop
7. no_of_purchased

if i want to find what impact to no_of_purchase and need to include both numeric and non-numberic data. which model and method should i use?

Answer 1

Since you want explainability of your feature parameteres, the simplest approach would be to use simple Linear Regression or Regression with handcrafted feature values. In this way, you'll get a weight associated with a each feature (may be positive or negative) which will tell you how exactly important it is. But before you actually implement a Linear Regression Model, you'll have to do some pre-processing by converting categorical features into their One-Hot Encoded form and hopefully normalize the continuous values.

If not linear regression, you could always go with ensemble methods (eg. RandomForest Classifier, XGBoost or LightGBM). They are very easy to use out of the box. They have an inbuilt feature importance metric present and use different criteria to calculate those metrics. You could go through their docs and see which one is more important to you.

More likely than not, ensemble methods would outperform the Linear Regression model, so in my opinion, they seem like your best bet.