How to find the classification Accuracy

Hello everyone.

I was working in below docuemntation, and I did run the code

https://www.mathworks.com/help/stats/optimize-an-svm-classifier-fit-using-bayesian-optimization.html

But I couldn't know where the accuracy exactly ? I need it in percentage.

Also how could I find where is the model accuracy for the training and the accuracy for the testing.

Would you please help.

Thank you

[Athul Prakash]

check the documentation for fitcsvm and you'll find your answers.

https://www.mathworks.com/help/stats/fitcsvm.html

Especially refer to the section: Input Arguments > Hyperparameter Optimization Options

Not sure which parameter exactly you wanted, but I think I found a way by loggin Misclassification rates. Method is outlined in my answer below.

Answer 1

[Athul Prakash]

Hi Noufal,

Here's how to log and MisclassificationRate and calculate Accuracy from there:

Where is the Misclassification Rate:

The 'objective Function' during minimization is actually the misclassification rate of the svm. These values are displayed in the output text, for each iteration separately.

To log those values:

By setting 'SaveIntermediateResults' property of 'opts' (HyperparameterOptimizationOptions) to true, fitcsvm will log information after every iteration - into a 'BayesianOptimization' object named 'BayesoptResults'.

After the process completes, examine the 'ObjectiveTrace' property of that object to view the record of objective function values for each iteration (these are the misclassification rates of the 30 different SVMs)

Percentage Accuracy

PercAcc = 100*(1-MisClassRate)

(I highly recommend going through the doc for Bayesian Optimization object - https://www.mathworks.com/help/stats/bayesianoptimization.html)

(Also check the 'Hyperparameter Optimization Options' section of the doc for fitcsvm function - https://www.mathworks.com/help/stats/fitcsvm.html)

Hope it Helps!