Earthquakes pose a significant threat to societies worldwide, underscoring the urgent need for advanced prediction technologies. This study introduces an optimization technique aimed at reducing the error rate in earthquake prediction by selecting the most suitable parameters for a Bi-LSTM (Bidirectional Long Short-Term Memory) model. Despite Bi-LSTM's promising outcomes, variations in parameters can impact performance, necessitating careful parameter selection. This research employs Grey Wolf Optimization (GWO) to optimize parameters and evaluates its effectiveness against other group optimization approaches to identify the most efficient parameters for earthquake prediction. Additionally, a multiple input multiple output (MIMO) architecture is implemented to enhance prediction accuracy. The evaluation results demonstrate that GWO outperforms other optimization techniques, achieving a reduced loss score of 0.364. The ANOVA method yields a p-value approaching 0, indicating statistical significance. This study contributes to the development of early warning systems for earthquake disasters by emphasizing the importance of parameter optimization in earthquake prediction and showcasing the effectiveness of Bi-LSTM and GWO methodologies.