You are tasked with developing a machine learning pipeline for training an XGBoost classification model using tabular data stored in a BigQuery table. Your goal is to ensure that the pipeline effectively handles data splitting, feature engineering, and model evaluation, and allows for easy comparison of different models. The required steps are: 1. Randomly split the data into training and evaluation datasets in a 65/35 ratio. 2. Conduct feature engineering to prepare the data for training. 3. Obtain evaluation metrics for the model's performance. 4. Compare the performance of models trained in different pipeline executions. Which approach should you take to achieve these requirements?