It is now time to finally draw our list, applying the majority vote technique we learned previously to our predictions. As done before, we are going to apply a threshold on values predicted from the logistic and SVM models, to map the original predictions on the [0,1] domain. Finally, with a piece of code really similar to the one we have seen before, let's create an ensemble_prediction
attribute, storing a final prediction defined from results coming from the three estimated models:
me_customer_list %>% mutate(logistic_threshold = case_when(as.numeric(logistic)>0.5 ~ 1, TRUE ~ 0), svm_threshold = case_when(as.numeric(svm)>0.5 ~ 1, TRUE ~ 0)) %>% mutate(ensemble_prediction = case_when(logistic_threshold+svm_threshold+ as.numeric(as.character(random_forest)) >=2 ~ 1, TRUE ~ 0)) -> me_customer_list_complete
Is this the list the internal audit team needs from us?
Not quite; there is one more computation required...