Application of artificial neural network with metaheuristic optimization for improving the nutritive value of fried fish

Nutritional quality indices of fish deteriorate drastically during frying. In this study, using Catla catla fish and mustard oil (culinary media), extensive experiments are carried out varying the temperature, time, and oil amount to attain the best nutritional quality indices of fried fish with a tuned combination of cooking parameters. An artificial neural network (ANN) is developed to select the best model to find a nonlinear correlation between the frying conditions and nutritional quality indices. ANN-based metaheuristic optimization methodologies, namely genetic algorithm (GA), differential evolution, firefly optimization, and gray wolf optimization (GWO), are applied to optimize the best cooking conditions. Among these, GWO is most promising for optimizing favorable inputs, practical optimal solutions, and reasonable execution time. As outputs are conflicting, multi-objective genetic algorithm (MOGA) is implemented for their simultaneous optimization with optimum values of process variables for health benefit, reducing frying time, and minimizing the wastage of culinary media. The MOGA successfully improves the omega-3/omega-6 fatty acids, polyunsaturated fatty acids/saturated fatty acids, cis/trans fatty acids ratio, and index of atherogenicity values up to 40.43%, 65.35%, 137%, and 83.84%, respectively, satisfying the multi-objective criteria. Practical applications The conventional frying process of fish is successfully optimized by the developed hybrid model-optimization topology that improves the nutrient value of fried fish significantly. The developed neural model automatically searches all available algorithms and activation functions exhaustively to select the best model. All single-objective and multi-objective genetic algorithms integrated with an artificial neural network can attain the optimum successfully for all outputs by a unique tuned cooking condition. Operators can choose the preferred solution among many Pareto optimal combinations as per their needs. Furthermore, this developed, generic topology provides a tool for process modification and optimization of other food process engineering methods.

Automated summary

This study successfully improves the nutritional value of fried fish by varying frying conditions and optimizing cooking parameters. The use of artificial neural network and genetic algorithm helps find the best frying conditions and optimization solutions, meeting the multi-objective criteria.