MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202641049484 A) filed by Keshav Memorial Institute Of Technology, Hyderabad, Telangana, on April 17, for 'hybrid stock market forecasting using transformers, reinforcement learning and sentiment-driven insights.'

Inventor(s) include Dr. Kumbham Bhargavi; Ms. Hema Sree Guddanti; Mr. Praneeth Venkata Reddy Kallam; Mr. Krishna Sai Gangishetty; Ms. Yadaboyina Rishitha; and Ms. Desetty Kavya Sri.

The application for the patent was published on May 1, under issue no. 18/2026.

According to the abstract released by the Intellectual Property India: "Predicting stock market movements remains one of the most challenging tasks in financial research due to high volatility, nonlinear dependencies, and sensitivity to external events. Traditional models such as ARIMA, LSTM, and GRU often capture only limited temporal patterns and fail to generalize well across varying market conditions. To address these challenges, in this paper proposed a hybrid forecasting framework that integrates Transformer-based deep learning, sentiment analysis of financial news, and Reinforcement Learning (RL) through Proximal Policy Optimization (PPO). The Transformer model captures long-range temporal dependencies in historical stock data, while sentiment analysis enhances predictive features by incorporating investor psychology and macroeconomic signals from news headlines. Our experiments show that the standalone Transformer model achieved a predictive accuracy of 52.88% on test data. These enriched features were then fed into a PPO agent, which dynamically learned optimal trading strategies under uncertain conditions. When integrated with reinforcement learning, the PPO agent significantly boosted performance, attaining 84% accuracy. These results highlight the potential of combining Transformers, sentiment-driven insights, and reinforcement learning for building robust and adaptive algorithmic trading systems that achieve higher predictive accuracy and better risk-adjusted profitability compared to conventional statistical and deep learning approaches."

Disclaimer: Curated by HT Syndication.