Fossil fuel-based power generation remains a core component of many energy systems. However, growing electricity demand influenced by government policies, economic development, and societal behavior can challenge the stability of power supply networks. This study focuses on improving the productivity of combined cycle power plants through machine learning techniques to better match electricity generation with consumption while minimizing inefficiencies. Several algorithms, including Bagging decision tree, Random Forest, Gradient Boosting, and XGBoost, are implemented to predict energy output. A parameter tuning process is conducted to enhance model accuracy using negative root mean square error as the evaluation metric. The results indicate that Gradient Boosting achieves the highest accuracy, with a mean absolute error of 2.215449. Additionally, a neural network model is developed to optimize key operational parameters such as temperature, ambient pressure, and humidity. This model integrates a Quasi-Newton-based activation function to improve adaptability and precision. The optimal values identified are a temperature of 19.38°C, exhaust vacuum of 25.43 cm Hg, ambient pressure of 1021.4 millibars, relative humidity of 60.8 percent, and an energy output of 462.09 megawatts. Temperature is recognized as the most significant influencing factor. Using real-world data from the Zavareh Combined Cycle Power Plant, the study offers practical insights for enhancing energy sector performance.