A Yield Prediction Model for Fintech Platforms Based on Investor Behavior Characteristics and Its Application in Capital Allocation Decisions

Investment decisions and capital allocation on fintech platforms often rely on experiential judgments, lacking systematic quantitative analysis of investor behavior changes. To reveal the impact of investor behavior on return formation mechanisms, this study constructs a "behavioral characteristics-return performance" prediction model based on 240,000 investor behavior data points from eight consecutive quarters on a fintech platform. This model evaluates the indirect effects of different capital allocation strategies on investment returns.The study first converts investor behavior into 12 reproducible feature variables, including position depth, trading frequency, cross-cycle revisit rate, and risk exposure adjustment behavior. It then compares the predictive performance of Lasso, Random Forest, CatBoost, and XGBoost models.Results indicate that incorporating behavioral features significantly enhances model interpretability, with the R² for return prediction increasing from 0.41 to 0.68. Among these features, "deep investment behavior" and "cross-cycle trading stability" contributed the most.Further scenario simulations reveal that increasing investment budgets alone does not directly boost returns, but significantly promotes deeper investor behavior. A near-linear relationship exists between depth behavior and return performance. This study uncovers behavioral transmission pathways through which capital allocation influences investment returns, providing a quantitative analytical framework for fintech platforms to make scientific investment and budget decisions.