I have a dataset that consists of a poisson distribution, a exponential distribution, categorical variables, and my target variable is a numerical bimodal variable. This is a regression model. I wanted to clarify the type of model best to use and any transformations required for my variables.
Model 1. Random Forest. Since RFs are scale invariant, no transformations are required for either target and independent variables.
Model 2. Gaussian Regression. Categorical Variables would be One Hot Encoded. The Poisson and Exponential Variables could be transformed with a Box-Cox Transformer. How would I transform my target bimodal variable? I already tried abs(y-mean(y))
, which when tested was not normal. Some things I have read, say to fit to a Gaussian Mixture Model, but I do not understand what they mean because I have only used GMMs with clustering.
Is there a different type of model that may work better here?