This study investigates the application of the ConcaveLSTM model, a novel machine learning approach combining the strengths of Stacked Long Short-Term Memory (LSTM) and Bidirectional LSTM, for predicting natural gas prices. Given the inherent volatility and complexity of energy markets, accurate forecasting models are crucial for effective decision-making. The research employs a comprehensive dataset from 1997 to 2020, focusing on the daily price of natural gas in US Dollars per Million British thermal units (Btu). Through rigorous testing across various model configurations, the study identifies optimal settings for the ConcaveLSTM model that significantly improve prediction accuracy. Specifically, configurations utilizing 50 input steps with neuron counts of 100 and 300 exhibit superior performance, as evidenced by lower Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE), alongside higher R-squared (R2) values. These findings validate the ConcaveLSTM model's potential in financial forecasting and highlight the importance of parameter tuning in enhancing model efficacy. Despite certain limitations regarding dataset scope and market variability, the results offer promising insights into developing advanced forecasting tools. Future research directions include expanding the dataset, incorporating additional market influencers, and conducting comparative analyses with other forecasting models. This study contributes to the evolving field of machine learning applications in financial market predictions, offering a foundation for further exploration and practical implementation in the energy sector.