This study uses a new Minimal Gated Unit (MGU) model to investigate stock price forecasting. It compares its effectiveness to that of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. The recurrent neural networks are implemented to develop a stock forecasting model that is more effective and precise, economical with various parameters, and reduces costs while increasing real-world viability during turbulent market situations. The methodology comprises dataset collection and preparation, model generation, hyperparameter optimization, and evaluation with RMSE, MAE, and MAPE. Results suggest that the RNN-MGU model is superior to both LSTM and GRU and confirms that the lowest RMSE is equal to those of other suggested models being 0.071884, MAE is 0.052852, and MAPE 0.100005 showing excellent prediction accuracy. Smart tuning of hyperparameters in Optuna is believed to be the key to better results, with neural units, learning rate, and dropout rate being the most critical.