I have worked in R previously but I'm quite new to predictive modeling with Random Forest and Gradient boosting models in R.
I have a dataset with 100 healthy individuals and 50% of the dataset has some degree of heredity for a autoimmune disease called systemic lupus erythematosus (SLE).
In this dataset there are approximately 200 predictors for each study participant and the predictors are quite mixed, ranging from demographics, to proteins and other molecules.
I have constructed and tuned Random Forest and Gradient boosting models for this dataset using a binomial variable called "Class" (i.e. No heredity vs Positive heredity) as the dependent variable. Both models shows very similar relative importance for predictor for the Class variable. After tuning the models I find that the predictor "hypertension" had the greatest relative importance to the models.
My question is the following, is it theoretically sound to create a predictive model with either RF or GBM to analyse the predictor "hypertension" to see which predictor out of the 200 possible, that has the greatest relative importance for hypertension?
Secondly, are there other methods to study "heredity" and it's influence on other predictors in such a dataset?
Thanks in advance!