<p>Predicting investments in dynamic stock markets is a difficult problem because of the high volatility, nonlinear price trends, dependency on time, and rapid change in market dynamics. Accurate forecasting is essential for supporting reliable investment decision-making, risk reduction, and financial market stability. However, existing stock prediction approaches often suffer from limited adaptability, overfitting, weak inter-feature relationship modeling, and high computational complexity under volatile market environments. To overcome these shortcomings, a new approach titled Dynamic Stock Market Investment Prediction is suggested based on Dual-Attention Similarity-Navigated Transformer Graph Network combined with Multiplayer Battle Game-Inspired Optimization (DASNT-GN-MBGO). The developed framework combines Correlation Coefficient based Min-Max Normalization (CC-MMN) as the preprocess step, Adaptive Causal Decision Transformers (ACDT) for temporal-causal feature extraction, and a combination of Dual-Attention Transformer Network with Similarity Navigated Graph Neural Network for sequential and relational learning. The MBGO technique was used further for optimization of network weights. Experiments were carried out on the Stock Market dataset with the same training settings. The results show that the framework presented in the paper demonstrated superior prediction accuracy with an MAE of 0.240, an RMSE of 0.310, an MAPE of 9.40% and R² equal to 98.85%. The proposed framework shows the high level of generalization, accurate forecasting abilities and efficient computations in the field of dynamic stock markets.</p>