Deep Learning and Optimization Techniques for Sentiment-Driven Stock Market Prediction Using Transformer Models

Stock price forecasting functions as a vital yet complex computational finance operation because financial markets demonstrate unpredictable and nonlinear patterns. This paper develops a framework for stock prediction that unites Transformer models with deep learning regression systems under metaheuristic optimization searches. DistilBERT performs embedding functions on financial news titles to extract sentiment information and combines them with market-based indicators for building comprehensive features. The fully connected neural network receives optimization from Optuna through Bayesian and random search algorithms to determine its hyperparameters. The analysis of Apple Inc. stock produced experimental results with an RMSE of 0.456 and an R2 score of 0.997. The proposed approach shows reliable performance according to visual and statistical evaluations. The research demonstrates that using market sentiment with data structures allows better market prediction which creates fresh opportunities to develop real-time explainable multi-asset financial models for the future.