Dear Jibril,

It is a nice question you asked.

The root relative squared error is relative to what it would have been if a simple predictor had been used. More specifically, this simple predictor is just the average of the actual values. Thus, the relative squared error takes the total squared error and normalizes it by dividing by the total squared error of the simple predictor. By taking the square root of the relative squared error one reduces the error to the same dimensions as the quantity being predicted [1]. You may also be interested to know the difference between high relative and root relative errors. To know this, refer to [2].

Reference:

[1] https://www.gepsoft.com/gxpt4kb/Chapter10/Section1/SS07.htm [2] https://stackoverflow.com/questions/36876749/high-relative-absolute-error-and-root-relative-squared-error-in-classification

Dear Jibril,

It is a nice question you asked.

The root relative squared error is relative to what it would have been if a simple predictor had been used. More specifically, this simple predictor is just the average of the actual values. Thus, the relative squared error takes the total squared error and normalizes it by dividing by the total squared error of the simple predictor. By taking the square root of the relative squared error one reduces the error to the same dimensions as the quantity being predicted [1]. You may also be interested to know the difference between high relative and root relative errors. To know this, refer to [2].

Reference:

[1] https://www.gepsoft.com/gxpt4kb/Chapter10/Section1/SS07.htm [2] https://stackoverflow.com/questions/36876749/high-relative-absolute-error-and-root-relative-squared-error-in-classification

Hi Muhammad,

The result of RRSE = 82.6872 % shows that your model is not quite good.

For more information, see this link:

https://www.gepsoft.com/gxpt4kb/Chapter10/Section1/SS07.htm

HTH.

Samer Sarsam, PhD.

Dear Sobhan,

Thanks

Dear Muhammad

82.6872 % is the accuracy level of the model that you have created. I do not think that it is a very good model as there are some model with accuracy as high as 99%. Classification is a method of supervised machine learning. There are various such type of methods. Here are some of the list of methods which you can try

1. Bayesian classification.

2. Support Vector Machines.

3. linear regression.

4. logistic regression.

5. naive Bayes.

6. linear discriminant analysis.

7. decision trees.

8. k-nearest neighbor algorithm.

9. Neural Networks (Multilayer perceptron)

The attached file is how to compute the root relative squared error.

To make things much clear, if the RRSE = 82.6872 % , then the model is not quite good. But if the accuracy = 82.6872 %, this means the model is arguably very good. Knowing that, in this case, looking at the accuracy alone is not enough, you have to explore another aspects like F-measure, Kappa statistic, etc.

Cheers,

Samer Sarsam, PhD.

Hi Sobhan Sarkar: Thank you for the answer! Can you please provide a reference for the "Relative Absolute Error" or the "Root Relative Squared Error"? A book or publication link would be really helpful (rather than stackoverflow).

Samer Sarsam

I join the answers with a new question:

What doas it mean if Relative absolute error is 71% and Root relative squared error is 78% ?

Accuracy is 58%.

Regards,

Željko Vujović

Zeljko Vujovic

Means the performance of the algorithm is not that good.

Cheers,

Dr. Samer Sarsam

Samer Sarsam

That is what I realized.

The questions are:

- What and how should it change?

- What and how much will the model learn from this dataset?

- Does any other algorithm preform better if it works with the same dataset?

- How can I find out?

Cheers,

Željko Vujović