In recent decades, the application of statistical methods an‎d machine learning algorithms in agricultural sciences has gained a prominent position as a powerful tool for enhancing productivity, improving decision-making, an‎d developing predictive models under variable climatic conditions. With the advancement of sensing technologies an‎d data collection methods, a massive volume of information has been generated from agricultural, environmental, an‎d biological experiments. Proper utilization of these datasets requires the application of rigorous scientific tools an‎d precise modeling approaches. In this context, improving an‎d optimizing the influential parameters of key agricultural processes plays a crucial role in fostering progress. In the present study, optimization of the experimental conditions for rice bran oil extraction was investigated with the aim of achieving the most favorable quantitative outcomes. Data related to three input factors, temperature, pressure, an‎d CO₂ molar concentration, were collected using a Central Composite Design (CCD). The corresponding outputs included oil yield, phosphorus content, an‎d free fatty acid concentration. Due to the complexity of the interactions among these variables, as well as the presence of nonlinear effects, the use of advanced statistical models an‎d machine learning techniques was deemed necessary. At the initial stage, the desirability function approach was employed to determine the optimal combination of input factors, successfully providing suitable parameter settings for the simultaneous maximization of multiple responses. Subsequently, in order to achieve more accurate modeling an‎d prediction, modern machine learning algorithms such as XGBoost, Support Vector Regression (SVR) with an RBF kernel, an‎d Gaussian Process Regression were implemented, along with fuzzy optimization techniques including Fuzzy Regression an‎d ANFIS. The results revealed that although XGBoost demonstrated acceptable accuracy for certain responses, it failed to provide a robust an‎d generalizable predictive pattern. In contrast, SVR with the RBF kernel exhibited superior performance, effectively capturing the complex nonlinear relationships between inputs an‎d outputs an‎d yielding the most accurate predictive results. Moreover, Gaussian Process Regression, owing to its probabilistic nature, not only achieved high predictive accuracy but also offered valuable insights into the uncertainty associated with predictions. Ultimately, it can be concluded that the choice of an optimization approach strongly depends on the final objective of the research. When resource an‎d time constraints are the primary considerations, the desirability function method serves as the most practical option. However, in scenarios where prediction accuracy an‎d the precise identification of variable interactions are of higher priority an‎d resource limitations are less critical, SVR with the RBF kernel emerges as the most effective strategy for optimizing the rice bran oil extraction process.

علي زينل همداني , علي اصغر بازدار

صبا صارمي نيا , مهدي خاشعي آشياني