linear regression, confidence limit and forward selection
Contributor I
Hello together,
I'm just beginning with rapid miner on some (for me) well known data, and several questions occur to me.
1. Assume numerical data with one label, on which a regression is performed. Is there a way to give a confidence or prediction interval for each predicted label? And if so, how can I get it?
2.
I am applying forward selection with linear regression as learner two times to a (purely numerical) example set. First time with the inner operators "linear regression -> apply model -> performance" as inner operators for the forward selection, and then with a X-validation in the forward selection. Inside the X-validation I have again linear regression as learner and apply model + performance as testing. The results in the performance vectors are slightly different as I expected, and each "method" produces an example set. The example set coming from the 'X-Validation-branch' does not contain predicted values in contrast to the one coming from the forward selection only with linear regression, although I would say, that the output of both coustructs in the forward selection is the same.
I can craete predicted values through another model application + performance after the X-Validation inside the forward selection, but I fear that this changes the result of the forward selection. What woud be a proper way to get predicted values of both methods?
I would highly appreciate if someone could give me a hint. Thank's in advance.
P.S. additionally the XML code:
<连接from_port = "训练" to_op = "线性回归sion (X-Val)" to_port="training set"/>
<参数键= " root_relative_squared_error”价值e="true"/>
<参数键= " root_relative_squared_error”价值e="true"/>
<连接from_port = "训练" to_op = "线性回归sion (2)" to_port="training set"/>
<连接from_op = "用" from_port = "输出1”to_op="Forward Selection (Regression)" to_port="example set"/>
I'm just beginning with rapid miner on some (for me) well known data, and several questions occur to me.
1. Assume numerical data with one label, on which a regression is performed. Is there a way to give a confidence or prediction interval for each predicted label? And if so, how can I get it?
2.
I am applying forward selection with linear regression as learner two times to a (purely numerical) example set. First time with the inner operators "linear regression -> apply model -> performance" as inner operators for the forward selection, and then with a X-validation in the forward selection. Inside the X-validation I have again linear regression as learner and apply model + performance as testing. The results in the performance vectors are slightly different as I expected, and each "method" produces an example set. The example set coming from the 'X-Validation-branch' does not contain predicted values in contrast to the one coming from the forward selection only with linear regression, although I would say, that the output of both coustructs in the forward selection is the same.
I can craete predicted values through another model application + performance after the X-Validation inside the forward selection, but I fear that this changes the result of the forward selection. What woud be a proper way to get predicted values of both methods?
I would highly appreciate if someone could give me a hint. Thank's in advance.
P.S. additionally the XML code:
<连接from_port = "训练" to_op = "线性回归sion (X-Val)" to_port="training set"/>
<参数键= " root_relative_squared_error”价值e="true"/>
<参数键= " root_relative_squared_error”价值e="true"/>
<连接from_port = "训练" to_op = "线性回归sion (2)" to_port="training set"/>
<连接from_op = "用" from_port = "输出1”to_op="Forward Selection (Regression)" to_port="example set"/>
Tagged:
0
Answers
there's currently no method available for the confidence or prediction interval modeling. In fact I don't know any method that will deliver this information...Does anyone?
To your second question:
The proper way is to use a X-Validation to estimate the performance. Otherwise you will test on the training exampleset. If you are going to get predictions after the forward selection, you will have to learn a model again and apply it separately on the subset that's the result of the forward selection.
Greetings,
Sebastian