Comparison of neural network and factorial design in optimizing red mulberry juice turbidity reduction

Background and Objectives: In recent years, increasing attention to the health and quality of food products has led to the importance of food processing using new methods. One of the significant challenges in this field is the turbidity of fruit juices, particularly red berry juice. An effective way to reduce turbidity is by using natural absorbents. In this regard, banana peel is introduced as a natural absorbent that can significantly contribute to the reduction of turbidity in red mulberry juice. Known for its rich compounds, including pectin and plant fibers, banana peel is recognized as a potential source for absorbing suspended particles and reducing turbidity.
Materials and Methods: The primary materials studied in this research are red berry juice and banana peel. The berry juice prepared for this research had a Brix value of approximately 78, which was reduced using distilled water through a series of 15 dilutions. The banana peel was dried at a mild temperature of 45 oC in a laboratory oven and then ground to a uniform size of 1 mm using industrial sieves. The study compared neural network and factorial methods for reducing the turbidity of red mulberry juice with the natural absorbent of banana peel. The independent variables considered included temperature, time, and stirrer speed, each investigated at four different levels, with the experimental design employing a factorial approach. The responses measured were the percentage reduction in turbidity and the percentage reduction in absorbent efficiency.
Results: The results showed that the maximum reduction in turbidity (47.22%) with the lowest reduction in the efficiency of the natural absorbent (57.65%) was achieved at a temperature of 30 oC, with surface absorption duration of 3 hours and a stirring speed of 300 rpm. Additionally, a neural network was utilized to predict the two dependent variables as functions of the independent variables. The neural network modeling demonstrated high accuracy in predicting the target variables, with mean relative error (MRE) values of 2.06% and 0.90% for the turbidity reduction percentage and the absorbent efficiency reduction percentage, respectively. In contrast, the factorial method yielded MRE values of 4.58% and 6.04%, thus significantly enhancing the prediction accuracy for the two dependent variables.
Conclusion: Banana peel, as a natural absorbent, was effective in reducing the turbidity of red mulberry juice. Furthermore, the factorial method proved to be effective as one of the experimental design approaches for identifying optimal operating conditions in the turbidity reduction process. The use of a neural network for predicting the results of laboratory research demonstrated a high degree of confidence in modeling outputs. In this study, the neural network provided improved predictions for the two dependent responses, leading to enhanced outcomes in both turbidity reduction and absorbent efficiency reduction.