Quantitative Assessment of Brix in Grafted Melon Cultivars: A Machine Learning and Regression-Based Approach.

Area of Science:

• Horticulture and Agronomy
• Machine Learning in Agriculture
• Plant Physiology

Background:

• Melon fruit quality is influenced by rootstock selection and biochemical composition.
• Accurate assessment of fruit quality traits like Brix content is crucial for agricultural management.
• Machine learning offers potential for non-destructive quality evaluation.

Purpose of the Study:

• To compare Support Vector Regression (SVR) and Multiple Linear Regression (MLR) models for predicting melon Brix content.
• To analyze the impact of different rootstocks on melon fruit biochemical properties.
• To establish a reliable, non-destructive method for assessing melon quality.

Main Methods:

• Grafting melon plants with different rootstocks (Sphinx, Albatros, Dinero).
• Measuring fruit biochemical parameters including nitrogen, phosphorus, potassium, calcium, magnesium, and Brix.
• Developing and evaluating SVR and MLR models using performance metrics (MAE, MAPE, MSE, RMSE, R²).

Main Results:

• The SVR model demonstrated superior accuracy in predicting Brix content, achieving an R² of 0.9904, significantly outperforming the MLR model (R²: 0.9472).
• Strong correlations were observed between Brix levels and sugar content (sucrose, glucose, fructose), with minimal influence from titratable acidity.
• SVR proved to be a more reliable and non-destructive method for melon quality assessment compared to MLR.

Conclusions:

• SVR is a highly effective machine learning tool for non-destructive melon quality assessment, particularly for Brix content prediction.
• Rootstock choice significantly impacts melon fruit biochemistry, which can be modeled using SVR.
• These findings support the optimization of rootstock management and cultivation practices to enhance melon fruit quality.