I am using SPSS 27. We did a COVID survey. I am trying to see the predictors of the 'perceived threat score' and used multiple factors like race, age, gender, financial status etc. as independent variable. I used Neural network Multilayer Perceptron. Saved the predicted 'perceived threat score'. However, I was very surprised to look into the output or new variable. Almost all the numbers are same, and ML model is much worse compared to general linear model. Not sure how to fix it. Attached a screenshot, observed and predicted columns are side by side (with predicted column having all same values).

There is also an error message- "One or more cases in the testing or holdout sample have factor or dependent variable values that do not occur in the training sample. These cases are excluded from the analysis."

FYI: We have 3000+ subjects.

Please help.

------------------------------
Pritish Mondal
------------------------------

#SPSSStatistics

I can't tell for sure what is wrong without seeing the data and specification, but the error message suggests a possible cause.  If some of the variables have values that did not occur in the training sample, they can't be used for prediction.  It looks like all the predictions are just using a mean or some other simple statistic.  If there are missing values, it may be using value substitution.  Take a look in the Scoring Wizard and make sure the variables line up.

------------------------------
Jon Peck
------------------------------

Original Message:
Sent: Thu December 31, 2020 11:01 PM
From: Pritish Mondal
Subject: Neural network Multilayer Perceptron

I am using SPSS 27. We did a COVID survey. I am trying to see the predictors of the 'perceived threat score' and used multiple factors like race, age, gender, financial status etc. as independent variable. I used Neural network Multilayer Perceptron. Saved the predicted 'perceived threat score'. However, I was very surprised to look into the output or new variable. Almost all the numbers are same, and ML model is much worse compared to general linear model. Not sure how to fix it. Attached a screenshot, observed and predicted columns are side by side (with predicted column having all same values).

There is also an error message- "One or more cases in the testing or holdout sample have factor or dependent variable values that do not occur in the training sample. These cases are excluded from the analysis."

FYI: We have 3000+ subjects.

Please help.

------------------------------
Pritish Mondal
------------------------------

#SPSSStatistics