Naïve Bayesian and the decision trees algorithms result comparison?

servicenowstar

I used cross-validation to compare the performance of the two models( Naïve Bayesian and the decision trees). Here is the result:

and 

As you can see above both models have the same overall accuracy. Class wise also the same but in different classes. Which one do you think has better performance?
Thank you in advance!!!

YYH

Hi @servicenowstar, 

thanks for sharing the findings. No single performance measurement is better. They are complementary to one another. In most of the application of machine learning models, difference performance corresponds to entirely separate strategic practices. 

Now it is a question to you. How do you accept the misclassification in difference classes? As you can see the limitations of classification accuracy and when it can hide important details. In business use cases, people usually define a value based performance measurement by combining the confusion matrix and a cost matrix to align the model performance with the business target by penalizing the misclassification of each class strategically. 

In AutoModel 9.4 Beta, you can edit the costs for wrong predictions after you select the target/label column



OK getting back to your original question, if a wrong classification of Virginia iris has higher cost than that cost of Versicolor iris, I would prefer to pick the model in your Figure 5.

Cheers,

YY

YYH

Hi @servicenowstar, 

thanks for sharing the findings. No single performance measurement is better. They are complementary to one another. In most of the application of machine learning models, difference performance corresponds to entirely separate strategic practices. 

Now it is a question to you. How do you accept the misclassification in difference classes? As you can see the limitations of classification accuracy and when it can hide important details. In business use cases, people usually define a value based performance measurement by combining the confusion matrix and a cost matrix to align the model performance with the business target by penalizing the misclassification of each class strategically. 

In AutoModel 9.4 Beta, you can edit the costs for wrong predictions after you select the target/label column



OK getting back to your original question, if a wrong classification of Virginia iris has higher cost than that cost of Versicolor iris, I would prefer to pick the model in your Figure 5.

Cheers,

YY