Hybrid Deterministic-Generative Framework: Integrating Financial Risk Classification and Narrative-Driven Portfolio Optimization

Traditional financial advisory systems often depend on unpersonalized rules and lack the dynamic adaptability required for modern portfolio management. To solve this problem, this study develops and evaluates a hybrid artificial intelligence framework that integrates machine learning with generative narratives. This model uses an XGBoost classifier optimized via the Optuna Tree-structured Parzen Estimator (TPE) and is evaluated using k-fold cross-validation (k = 5). The optimized model achieves an overall accuracy of 90.90% in classifying personalized financial risk profiles. Furthermore, a dynamic allocation mechanism interprets these probabilistic predictions into risk-adjusted portfolio distributions by leveraging real-time financial APIs. Moreover, this framework integrates the locally deployed Qwen 2.5 (7B) Large Language Model (LLM) to generate qualitative financial justifications. The LLM generates personalized recommendation narratives based on the prediction labels produced by the previous model. Therefore, this integrated framework provides a solution to the limitations of conventional financial recommendation systems while offering real-time and data-driven adaptive financial recommendation. Accepted at ICoDiT 2026.

Recommended citation: Wijaya, D., Deviana, F., Jonowidjojo, G., Yap, S. S., Sihombing, T. S., & Tarigan, G. A. (2026). "Hybrid Deterministic-Generative Framework: Integrating Financial Risk Classification and Narrative-Driven Portfolio Optimization." The 1st International Conference on Computing and Digital Technology (ICoDiT).