<p>This study investigates the critical but often overlooked role of distance metric selection in classification models using mixed data, with a focus on P2P loan performance forecasting. Unlike previous studies that used standard distance metrics with minimal critical evaluation, we systematically evaluate 24 distance metrics across four diverse P2P lending platforms. Our results demonstrate that a simple mixed-data distance metric significantly improves prediction accuracy and computational efficiency, while effectively capturing loan dependencies. Furthermore, the appropriate choice of distance metric enables investors to filter underperforming loans more effectively, optimizing portfolio outcomes. This research provides a robust methodological framework for enhancing credit and profit scoring models in P2P lending, providing investors with a robust, scalable, and efficient analytical tool. </p>