• Version
• Download 6
• File Size 382.48 KB
• File Count 1
• Create Date 22 March 2026
• Last Updated 22 March 2026

A Hybrid Machine Learning and Generative AI Architecture for Probabilistic Stock Range Prediction

R. SRINIKETH

Assistant Professor, Dept of CSE CMR Technical Campus Hyderabad, Telangana, India sriniketh@gmail.com

B. VAISHNAVI

UG Student, Dept of CSE CMR Technical Campus Hyderabad, Telangana, India 237r1a0571@cmrtc.ac.in

T. SAATHVIK REDDY

UG Student, Dept of CSE CMR Technical Campus Hyderabad, Telangana, India 237r1a05c1@cmrtc.ac.in

M. PRASHANTH KUMAR

UG Student, Dept of CSE CMR Technical Campus Hyderabad, Telangana, India 237r1a0599@cmrtc.ac.in

Abstract—In the financial sector, stock market prediction is traditionally approached by attempting to forecast a singular, exact future price point. However, these deterministic models frequently fail due to the inherent volatility of global markets. This paper presents a novel, web-based financial architecture that combines probabilistic machine learning with a Generative Artificial Intelligence (AI) translation layer. Instead of predicting an exact closing price, the proposed system utilizes XGBoost configured with Quantile Regression to generate a 90% confi- dence ”Safety Zone” (bounded by the 5th and 95th percentiles). To bridge the gap between complex quantitative outputs and retail investor comprehension, the architecture integrates a Large Language Model (LLM), specifically Gemini Flash, to dynami- cally translate technical indicators and real-time news sentiment into plain-English analogies. Deployed via a Streamlit interface, empirical backtesting over a 30-day trailing window demon- strates that bounding forecasts within dynamically adjusting quantiles significantly improves mathematical reliability while democratizing financial analytics for non-institutional users.Index Terms—Stock Market Prediction, Machine Learning, XGBoost, Quantile Regression, Generative AI, Large Language Models, Technical Analysis