<p>Neural networks are widely utilized in commerce, science, engineering, and medicine, owing to their versatile applications. However, optimizing their performance requires meticulous parameter design, which presents a significant challenge. The existing literature frequently uses labor-intensive trial-and-error methodologies to set parameters, considering critical factors such as momentum, neurons, transfer function, training, and learning rate, all of which significantly influence model accuracy. To address this challenge, this study proposes a novel approach using a modified version of the Taguchi experimental technique to determine the optimal parameter configuration for a feed forwardback propagation-trained neural network. This paper includes a case study focused on forecasting the specific fuel consumption of diesel engines fueled with various composition plastic pyrolysis oil and diesel, thereby demonstrating the effectiveness of our approach. The most efficient neural network parameter values were identified through rigorous experimentation and statistical analysis. The results of this research underscore the superior performance of the neural network optimized using the Taguchi technique compared with models with randomly selected parameters. This underscores the significance of our approach for enhancing the predictive accuracy and efficiency of neural network applications.</p>