increase positive predictive value in xgboost binary classification

Question

I'm working on a binary classification project in an insurance company.

the dataset has 70 features including BMI, jobs, etc, and about 35,000 rows.

Based on the binary classification score (the probability value multiplied by 100) lower than 1, we accept the contract. (e.g., if the original probability exceeds 0.01, the contract falls into a rejection region.)

my goal is to improve accuracy (to bring more people automatically in) and improve positive predictive value at the same time. (in this case, the positive value is 0, acceptance of the contract). I know this sounds greedy but if there's any way to improve it, I'd like to take any advice.

so far out of 1259 test cases

 xgb hyperparameters
 : eta = 0.5, gamma = 0.25, max_depth = 25, subsample = 1, colsample_bytree = 0.6
   (definitely gonna overfit, but it gives me the best result in terms of positive predictive value)
 

 cancer_test$test_YN = as.factor(ifelse(test_prediction < 0.01, 0, 1)
 confusionMatrix(cancer_test$test_YN, test_label, positive = '0')

              reference
              0     1
 prediction
          0   538   15
          1   585   121


Pos Pred Value = 0.9729
Sensitivity    = 0.4791
#(these are the two values I wanna improve)

Answer 1

等待大神答复