Applies to:  NeuralTools, all releases

 

Variable Correlations and Interactions in NeuralTools

 

Dimension reduction / combining correlated variables:

• NeuralTools does not perform dimension reduction or combine correlated variables.

• Correlations are generally less of a problem for neural networks than for linear regression:

  • MLF nets handle even large amounts of correlation without issue.

  • PN/GRN nets can be affected if groups of redundant variables dominate the decision process.

• Recommended practice:

  • Check for correlations using StatTools Correlation and Covariance.

  • If correlations are significant, MLF nets are usually preferred. PN nets may still work if testing accuracy is good.

Variable interactions:

• Interactions (effects of one variable depending on another) are not a problem for neural networks.

• NeuralTools does not provide specific tools or panel graphs for analyzing interactions.

Summary: NeuralTools focuses on prediction accuracy rather than simplifying correlated variables or visualizing interactions. Users can use external tools (e.g., StatTools) to explore correlations.