"why we dont see accuracy rate for MLP and help about RMSE"

hakmesyo

Hi,

I am new user of RapidMiner. I have a couple of questions.

First: I can see accuracy rate for some learner (i.e. tree induction learner) but in the case of ANN learner we just see RMSE error
Second : at the result of Neural net learner if we have a "root_mean_squared_error: 0.452 +/- 0.463 (mikro: 0.647 +/- 0.000)".  For the value , what does it mean? is it good value for class seperability or not. For example when i use SVM it give me "root_mean_squared_error: 1.036 +/- 1.343 (mikro: 1.696 +/- 0.000)". Could you give detailed explanations about RMSE message. And make comparison between this two leaner based on RMSE results.

Thanks for your helps and advice 

land

Hi,
it seems to me, that you are performing a regression task on classification data. If you have data with a label, the data type of the label attribute determines if a regression of classification is performed. The Root Mean Squared Error is a measure for numerical deviation from the true value, while the accuracy measures the percentage of correctly classified examples. For more details on Root mean squared error I would suggest taking a look in wikipedia.
The meaning of micro and macro average is explained in some other threads of this forum.

Greetings,
  Sebastian

hakmesyo

Thanks for your explanations. Actually I just wonder, getting RMSE error for example 0,15 does it mean 15 percent error or another meaning?

land

Hi,
for details about the root mean squared error, please refer to any book or other source about regression. It's probably the most basic error measure possible and is widely used. So it should be no problem to find anything about it on google.

Greetings,
  Sebastian