Fruit size prediction of tomato cultivars using machine learning algorithms.

Area of Science:

• Horticultural Science
• Agricultural Engineering
• Data Science

Background:

• Accurate early prediction of tomato fruit size is vital for greenhouse cultivation management and supply chain efficiency.
• Reducing the yield of small-sized tomatoes is a key objective for growers.
• Machine learning offers potential for developing predictive models in agriculture.

Purpose of the Study:

• To develop and evaluate machine learning models for early prediction of tomato fruit size at harvest.
• To compare the performance of Ridge Regression, Extra Tree Regression, and CatBoost Regression models.
• To assess the impact of incorporating average temperature on prediction accuracy.

Main Methods:

• Utilized fruit diameter data over time and cumulative temperature after anthesis to estimate fruit weight.
• Trained and evaluated three machine learning models (Ridge Regression, Extra Tree Regression, CatBoost Regression) using PyCaret.
• Tested models on three tomato cultivars ('CF Momotaro York,' 'Zayda,' 'Adventure') using different prediction periods.

Main Results:

• Ridge Regression achieved the lowest mean absolute percentage error (MAPE) of 9.8% for 'Zayda' using data at 200°C d, 300°C d, and 500°C d.
• Extended prediction periods (300°C d, 500°C d, 800°C d) improved accuracy for all cultivars with Ridge Regression (e.g., 8.8% for 'Zayda').
• Adding average temperature during the prediction period slightly enhanced model performance.

Conclusions:

• Machine learning models, particularly Ridge Regression, can effectively predict greenhouse tomato size at harvest.
• Early fruit size prediction aids in cultivation management, such as fruit thinning.
• Automating fruit diameter data acquisition could further enhance the utility of these predictive models.