Grape vine (Vitis vinifera) yield prediction using optimized weighted ensemble machine learning approach

Abstract

Grape vine (Vitis vinifera) plays a significant role in the agricultural industry, contributing substantially to the global economy through the production of table grapes, wine, and raisins. With increasing demand for high-quality grapes, both for domestic consumption and export, there is a pressing need to improve yield prediction models for better resource management. In this study, we propose an optimized weighted ensemble machine learning approach for predicting grape vine yield, integrating multiple morphological, physiological, and berry quality parameters. A diverse set of machine learning (ML) models, including Random Forest (RF), Artificial Neural Network (ANN), Extreme Gradient Boosting (XgBoost), Support Vector Regression (SVR), Gaussian Process Regression (GPR), Cubist and Multivariate Adaptive Regression Splines (MARS), were employed to model the grapevine yield. A Minimum Data Set (MDS) selection was performed using Principal Component Analysis (PCA), followed by data normalization to enhance model efficiency. Additionally, three ensemble approaches-Simple Averaging, Weighted Averaging, and Ridge Regression-based ensemble models were implemented to improve prediction accuracy. The dataset was divided into training and testing subsets, with hyperparameters of each model tuned using repeated k-fold cross-validation. The ensemble approach demonstrated superior performance, with improved accuracy in yield prediction compared to individual base models. This study highlights the effectiveness of ensemble learning in precision viticulture, offering a reliable framework for yield prediction in grapevine cultivation. The proposed approach offers a practical framework for vineyard managers and growers to optimize resource allocation and improve decision-making.