Machine Learning-Driven Analysis of the Impact on Risk Preferences of Peer-to-Peer Investors in the United States: Incorporating the Effects of Natural Disasters

This paper uses the lending transactions of the US Prosper platform from 2005 to 2008, combined with the natural disaster records of various states in the United States, and adopts machine learning techniques such as random forests and XGBoost to analyze the impact of these internal and external factors on investment decisions, focusing on the impact of natural disasters on the risk preferences of P2P investors. The analysis shows that natural disasters significantly change investors' risk preferences, while factors such as financing ratio and borrower income play an important role in investor decision-making. The results emphasize that P2P lending platforms during disasters need strong risk management strategies to stabilize market sentiment and investor behavior.