Efficient Tabular Data Preprocessing of ML Pipelines