Approach – 2 :
  Random forest gives us variable importance. Using these important variables  models are built in this approach.
         Logistic Regression:
         XG Boost:  
   
Model
Accuracy
Precision
F1 – score
Recall
Logistic Regression
48
51
48
48
XG - Boost
59
71
63
59
 
Approach – 3 :
  In this approach, target variable is made to binary class from multi-class
             Logistic Regression:
             Random Forest:
         
Model
Accuracy
Precision
F1 – score
Recall
Logistic Regression
60
62
59
61
Random Forest
62
60
61
62
 
              Logistic Regression:  
             Random Forest:
Model
Accuracy
Precision
F1 – score
Recall
Logistic Regression
76
93
86
76
Random Forest
77
95
87
77
 

Conclusion:

Recall is the error metric,  we cannot classify a patient wrongly as he cannot be readmitted (treatment should be done on time, this may also cost a patient's life).
Recall = True Positive / Total Actual Positive 
Based on the approaches made during model building, it can be concluded that these models can be used by the client to predict hospital readmission