Regularized discriminant analysis

Saad

Hi there !

I have a dataset (csv format) on which I want to run the Regularized discriminant analysis but I am getting this error mesage;

"The operator RDA does not have sufficient capabilities for the given data set; polynomial attributes not supported"

Can anyone help me with this?

Thanks!

varunm1

Hello @Saad

The error implies that this particular operator cannot handle categorical variables (polynomial type). Each operator (algorithm) has some restrictions on what kind of data they can handle. I see that you need to provide only "Numerical attributes" (Predictors/variables) for this model. 

To check what kind of data types you can use. You just need to right-click on the operator and click "Show Operator info".

Let us know if you need more information.

Saad

I tried running the model by turning the binomial variables in to 1 and 0 and turning the attribute to integer. The model did run but the results were not good. It was predciting with 50% accuracy for one class and showing 0% for another class

varunm1

Hello @Saad

There can be multiple reasons an algorithm cannot perform well on data. You need to do data preprocessing (Feature selection, class imbalances, etc.) before applying your model, optimize model by tuning hyperparameters and the type of validation you are using is also important. Also, this algorithm might not be good for the data you are trying to fit.

If you are looking for some resources that help you understand modeling, you can check the academy.rapidminer.com 

Saad

Can you please guide me regarding the pre-processing?

jacobcybulski

Hi @Saad , I agree with @varunm1 that you need to work on your data pre-processing. However, I wonder if you have tried using a different classifier first? If so, are you getting a better or worse result? Have you tried applying a linear discriminant analysis or quadratic discriminant analysis first? Regularized discriminant analysis is the "in-between" model. One reason you may be getting very poor results with regularized classifiers is because you need to adjust the regularization parameter "alpha". The problem is that you cannot guess it but you need to experiment with it. Discriminant operators tend to overfit data, so you can use regularization to relax the fit so that the model performs better on data not seen before, such as a validation data set. I suggest trying a few different values for the alpha parameter between 0 and 1. @varunm1 suggested tuning the hyperparameters, which is a more advanced search for the best parameter using one of the optimization operators, e.g. "Optimize Parameters (Grid)" which comes with its own tutorial on how to use it. If RDA gives you a bad result for all regularizations, you may be using the wrong approach to classification, may be try a decision tree first?

Jacob