Currently, NaN values in tabular datasets are replaced with 0. Instead, we can check for NaN values at the data connection stage and raise an error.