topic Re: SAS ENTERPRISE MINER - Decrease Misclassification in Neural Network in SAS Data Science

SAS ENTERPRISE MINER - Decrease Misclassification in Neural Network

Jph1030 — Thu, 08 Dec 2022 15:46:16 GMT

Hello!

This probably sounds fairly simple, but I am brand new to this, does anyone have tips on how to simply reduce the misclassification rate when using a neural network node? Thank you!

Re: SAS ENTERPRISE MINER - Decrease Misclassification in Neural Network

ger15xxhcker — Fri, 09 Dec 2022 21:05:25 GMT

There are several ways to reduce the misclassification rate when using a neural network node in SAS Enterprise Miner, including:

1. Increase the number of layers and nodes in the neural network.
2. Increase the amount of training data or reduce overfitting.
3. Perform feature selection or scaling of features.
4. Add regularization to the neural network architecture to reduce overfitting.
5. Adjust learning parameters such as the learning rate.
6. Experiment with different activation functions.
7. Increase or decrease the number of iterations used to train the model.
8. Use different optimizers such as Adam optimizer.
9. Try ensemble methods such as bagging or boosting.

Re: SAS ENTERPRISE MINER - Decrease Misclassification in Neural Network

sbxkoenk — Wed, 14 Dec 2022 21:16:58 GMT

Hello,

It's better to give a bit more info!

That makes it easier for us to give a focused answer.

Do you have a binary target or a multi-class (multinomial) target variable?
Is the event of interest a rare level?
Do you have an abundance of observations such that you can do data splitting? (there are alternatives if not)
What's the dimensionality in your input space (number of x-variables)? Ever heard about dimensionality reduction or feature engineering?
Are you running into overfitting problems (like good results on TRAIN data , but poor results on VALID data)?
...

Koen