Determining Optimal Quality Parameters to Predict Mushroom Strains Using Polynomial Networks and Backpropagation

Mushrooms are extensively cultivated across the world and are highly valued for their notable nutritional benefits and health properties, while also being highly versatile and profitable. Consequently, many works have studied the nutritional parameters that define the quality of this food, although not all such parameters may be significantly representative of mushroom quality. The aim of the present work is thus to generate a novel model that allows mushroom strains to be predicted, using and integrating group method of data handling (GMDH) networks with backpropagation and widely used optimization algorithms, such as adaptive moment estimation (ADAM), for the optimal and accurate selection of combinations of quality parameters of mushroom strains in order to obtain greater profits. For this purpose, we used 7 mushroom strains ( Agaricus bisporus ) characterized by 27 quality parameters including texture and colour. The models implemented yielded positive results, establishing positive synergies and enhancing the predictive effect of mushroom strains. A reliability of 91.5% was obtained with the optimal combination of 8 quality parameters and a reliability of 92.5% was achieved through the optimal combination of 12 quality parameters. Thus, this study demonstrates the high reliability of the model generated for predicting representative quality parameters in button mushrooms using GMDH-ADAM, which may also represent a methodological advance for future research with other types of crops.