Robust Estimation of Wild Blueberry Yield via Triple Integrated Machine Learning Techniques

Abstract

Vaccinium angustifolium, commonly referred to as the wild blueberry, is a wild species native to North America and valued for both its taste and nutritional benefits. Packed with antioxidants, they further provide a variety of health benefits and help maintain biodiversity when grown within their natural habitats. Wild blueberries are considered a kind of “superfood” due to their power-packed nutrition. Wild blueberries are of crucial economic importance because they provide employment opportunities and open new markets for exportation. Machine Learning (ML) is used to model and analyze a number of parameters influencing crop production, and this makes it a major asset in the prediction of wild blueberry output. ML methods like Gaussian Process Regression Networks (GPR) Networks and Histogram Gradient Boosting Regression (HGBR) are used for this purpose. These models, in turn, predict wild blueberry production using the Marine Predators Algorithm (MPA) and Northern Goshawk Optimization (NGO). The corresponding models and optimizers are put into place to improve the accuracy of the results. Among them, HGNG is the best performer, with an R² of 0.996 in the remarkable training phase. After that, in this respect, HGMP performed well, which has an R² of 0.985 at this phase. Moreover, regarding wild blueberry yield forecasting, HGB did very well and had an R² of 0.976 during the training process.