Use a Random Forest Regressor, as this makes allowances for the heavy nonlinearities involved in mixed discrete and continuous data. Do k-fold cross validation, take the 90% confidence interval for the RMSE's, and see if they are below the threshold.