The emergence of online lending services such as peer-to-peer (P2P) lending has simplified transactions for lenders, eliminating the need for traditional bank intermediaries. However, accurately predicting potential lender defaults is a critical challenge in preventing financial distress, as lenders bear the burden of default risk. This challenge becomes even more complicate with P2P lending datasets that feature a multitude of complex attributes. To improve predictive accuracy, the study focuses on optimizing feature selection in this data-rich environment. Its main objective is to identify the most influential features for predicting default risk among P2P lenders through feature optimization techniques, leveraging the Genetic Algorithm (GA) in conjunction with ten different Machine Learning models. The study employed a hybrid approach of GA with three mutation rate levels, basic (MR=0), moderate (MR=0.5), and extreme (MR=1), to provide insights into model responsiveness and performance under various mutation scenarios. Notably, the results highlight GA+XGBoost as the best-performing model with a stable fitness score of 86.132% compared to others. This research holds significant potential for improving lender risk management in the P2P lending landscape by effectively identifying higher risk lenders. Ultimately, the findings contribute positively to the mitigation of financial risks for lenders within the P2P lending ecosystem.