A Hybrid Investment Risk Prediction Framework Integrating ADASYN-RF, CS-SVM, PCA-BP, and ARIMA Models

Abstract

Investment risk is often the result of a long-term interplay among multiple factors, such as market volatility and business performance. Predicting investment risk in advance helps investors avoid losses and protect their assets. However, current prediction methods rely heavily on historical data, leading to poor timeliness and unreliable results. Therefore, this study proposes an investment risk prediction model based on the adaptive comprehensive sampling algorithm and the random forest algorithm to predict investment risks accurately. This model makes full use of the adaptive comprehensive sampling algorithm to balance the categories of risk data, and the random forest captures the risk characteristics to achieve investment risk prediction. CS-SVM is introduced to improve the prediction model and prevent overfitting. Meanwhile, principal component analysis and backpropagation networks are combined to solve the problems of unstructured data processing and time series dependency modeling. In the experiment, the study took historical investment data as the dataset and compared it with the investment risk prediction models constructed by three algorithms: CatBoost-GJO, TS-GA, and GAN-Stacking. Evaluate the prediction accuracy of each method in investment risk prediction, including the Sharpe ratio, value at risk, volatility, win rate and profit and loss. The results show that the prediction accuracy of this model reaches 99.0%, the Sharpe ratio is 1.9, and the maximum drawdown is 5.9%, all superior to the comparison models. Moreover, in the actual investment risk prediction, its volatility and value at risk are only 8.2% and 7.3% respectively, while the winning rate and profit and loss ratio can reach 88.9%, 6:1. These results indicate that the proposed model achieves high accuracy in investment risk prediction, effectively addressing the limitations of existing methods and providing a new approach to improve prediction performance. It contributes to the development of intelligent and efficient investment risk forecasting.