Hello Sobhan,

Evaluating a classifiers performance, both scores (accuracy (alpha) and the root mean squared error (RMSE)) can be calculated from the true positive rate (TP), true negative rate (TN), false positive rate (FP) and the false negative rate (FN) in the case of a binary problem. The total size of the test dataset is Omega=TP+TN+FP+FN. The accuracy alpha is alpha=(TP+TN)/Omega and RMSE=sqrt((FP+FN)/Omega).

Consequently, the relationship is alpha+RMSE^2=1.

However, the RMSE is not very commonly used to evaluate classification performance, it is rather used for regression (continuous variables). There is a nice overview of performance scores on wikipedia https://en.wikipedia.org/wiki/Evaluation_of_binary_classifiers which also includes the influence of an unbalanced test dataset.

Regards,

Michael

RMS error is the function of difference between the True and Predicted target outputs. During the training stage of classifier, we need to train the system with the minimum RMS error (Error Threshold). If the system error converged below to minimum error threshold, then the accuracy of the system during the testing stage will be high. Sometime if the error is not converged below the error threshold, then programmer may stop the training phase based on iteration level, but this strategy may lead to less accuracy.

Hello Sobhan,

Evaluating a classifiers performance, both scores (accuracy (alpha) and the root mean squared error (RMSE)) can be calculated from the true positive rate (TP), true negative rate (TN), false positive rate (FP) and the false negative rate (FN) in the case of a binary problem. The total size of the test dataset is Omega=TP+TN+FP+FN. The accuracy alpha is alpha=(TP+TN)/Omega and RMSE=sqrt((FP+FN)/Omega).

Consequently, the relationship is alpha+RMSE^2=1.

However, the RMSE is not very commonly used to evaluate classification performance, it is rather used for regression (continuous variables). There is a nice overview of performance scores on wikipedia https://en.wikipedia.org/wiki/Evaluation_of_binary_classifiers which also includes the influence of an unbalanced test dataset.

Regards,

Michael

"Accuracy of a classifier" needs to be quantified; either using Square errors, likelihood, or any other meaningful metric that the designer picks that measure the distance between model-output and data-label.

The Root Mean Square Error (RMSE) is a frequently-used measure of the differences between values predicted by a model or an estimator and the values actually observed from the thing being modelled or estimated for SVM or ANN

So, Small RMSE means good learning and high accuracy results.

thanks.

RMSE - is the error measured between the actual model output and desired model output.

Accuracy of a classifier that is recognition accuracy depends on how many number of test patterns does exactly classified to their original pattern actual class.

Recog.Accuracy =100x (patterns classified correctly)/Total number of test patterns