Artificial Intelligence Models for Predicting Root Traits of Chokeberry Under Salt Stress

Abstract

Chokeberry (Aronia melanocarpa) is a recently introduced functional berry in Türkiye. It has a high health-promoting potential and growing commercial value. However, limited information is available regarding its physiological responses to abiotic stresses such as salinity. This study aimed to investigate the effects of salt stress on the root architecture of chokeberry plants grown in different growing media (soil and peat) and irrigated with five different salinity levels (0.65-10 dS m⁻¹). Root traits including fresh and dry weight, total root length, surface area, volume, average diameter, number of tips, forks, and crossings were measured using WinRhizo software. Additionally, the study employed machine learning algorithms XGBoost, Multilayer Perceptron (MLP), and Gaussian Process Regression (GPR) to predict root traits based on salinity levels and identify the most accurate predictive model. The results showed that increasing salinity significantly reduced all root growth parameters. Among the tested models, XGBoost achieved the highest predictive performance (R² > 0.9), followed by MLP and GPR. Fresh and dry root weights were predicted with 98% and 97-98% accuracy, respectively, while MLP was most effective in estimating surface area and root tips. However, predictions for average diameter, root volume, and root crossings showed lower accuracy (MAPE > 10%). The findings indicate that artificial intelligence-based models can successfully estimate chokeberry root responses to salt stress and offer a powerful tool for sustainable cultivation.