S. A. Shaik Mazhar 1 D. Akila

Balancing of DATA 
When designing the predictive model, the final consideration is the frequency of results and when any 
changes in data are necessary to take care of the real objective result of the interest. Because deaths may have 
a major economic impact, considering their frequency, predictive modeling also focuses on unusual effects 
(e.g., death). Several analytical instruments can be helpful in developing models with unusual results. 
The region under the receiving feature curve can be used to compare models to determine If there are 
differences in accuracy of the final classification or relative improvements over the normal target variable 

---

Nat.Volatiles&Essent.Oils,2021;8(5):5393-5404
 
5398 
occurrence rate. In the same set of classification algorithms, mortality estimates were produced (Bayesian 
network, decision stump, filtered classifier, boosted logistic regression, logistic regression, multi-boosted 
regression, naive Bayesian, random forest and voted perceptron). When developing predictive models for 
unusual events, it is important to balance training data dependent on the probability of incident occurrence.