Applies to: NeuralTools, all releases

For categorical prediction, does NeuralTools use a probit, logit, or multinomial function?

 

Answer

No. NeuralTools uses Probabilistic Neural Networks (PNNs) rather than traditional logistic/probit models. PNNs estimate a probability density function for each category based on the historical data. Probabilities for a new case come from interpolating across all historical data using smoothing factors. Unlike logistic regression, which compresses the data into a simple function, PNNs retain all information, often yielding more accurate predictions.

 

Related Article:

Technical Questions about the Training Process