Explaining black boxes

Using SHAP (SHapely Additive exPlanation)

Random Forest

• Overall weight
  
  

Shap values of all the records representing which variable play what kind of role in predicting the output. Eg: Sex when High (Male) playe a negative role in survival but female have very high chances of survival spread across other factors too

• Per prediction
  

Using eli5 (Explain like I'm 5)

Logistic Regression

• Overall weight
  
  
• Per prediction
  

Random Forest

• Overall weight
  
  
• Per prediction
  

Lime

• Per prediction as it is local