Our World Statistics Day conversations have been a great reminder of how much statistics can inform our lives. Do you have an example of how statistics has made a difference in your life? Share your story with the Community!
Choose Language Hide Translation Bar
Highlighted

## Question of missing informative in Neural Networks

I'm using neural networks in JMP pro and have a question about missing informative in it.

JMP manual says

'For a continuous variable, missing values are replaced by the mean of the variable. Also, a missing value indicator, named <colname> Is Missing, is created and included in the model.' but what exactly does it mean "a missing value indicator, named <colname> Is Missing, is created and included in the model". please teach me mathematical procedure if someone knows it.

Thank you,

1 ACCEPTED SOLUTION

Accepted Solutions
Highlighted

## Re: Question of missing informative in Neural Networks

The values are the weights applied in the linear combination for a node or the response if the term is an input or a node, respectively. So 0.395636947029777 is the weight for X2. It contributes 0 when X1 is not missing and 0.395636947029777 when X1 is missing.

The value 47.2628571428571 is the constant in the linear combination of the hidden layer. It is not the mean, and it depends on missing rows.

Learn it once, use it forever!
4 REPLIES 4
Highlighted

## Re: Question of missing informative in Neural Networks

The indicator column in the model (not the data table) contains 0 if the observation is not missing and 1 if it is missing. This indicator variable becomes another predictor. The weights for the indicator are close to zero if missing data is not informative.

Learn it once, use it forever!
Highlighted

## Re: Question of missing informative in Neural Networks

I could understand 0 or 1 chosen algorism when mising data exists.

Still have a question, following formulation is a part of the script linking between one Xfactor and first node in hidden layer.

(-0.0023784672793723)*X1+X2*(0.395636947029777+(-0.0023784672793723)*47.2628571428571)

X1=not missing value, X2=0 when not missing, 1 when missing,

-0.0023784672793723 looks weights.

I wonder what 0.395636947029777 and 47.2628571428571 are??

mean is 47.9287 in this case not 47.2628571428571...

Highlighted

## Re: Question of missing informative in Neural Networks

The values are the weights applied in the linear combination for a node or the response if the term is an input or a node, respectively. So 0.395636947029777 is the weight for X2. It contributes 0 when X1 is not missing and 0.395636947029777 when X1 is missing.

The value 47.2628571428571 is the constant in the linear combination of the hidden layer. It is not the mean, and it depends on missing rows.

Learn it once, use it forever!
Highlighted

## Re: Question of missing informative in Neural Networks

Oh, The value 47.2628571428571 is the constant that is culculated!

I got it, really appriciate it !!!

Could you tell me the name of this method?? very intersting.

Another question is why 0.5 is multiplied when tanh function is chosen as activation function??

The script says 'tanh 0.5*((-0.0023784672793723)*X1+X2*(0.395636947029777+(-0.0023784672793723)*47.2628571428571))

X1=not missing value, X2=0 when not missing, 1 when missing,

Article Labels