Machine Learning Models Applied to Predict Crop Yield of Loquat Trees (Eriobotrya japonica Lindl.)
摘要
Loquat (Eriobotrya japonica Lindl.) is a subtropical fruit tree gaining popularity nowadays due to its higher nutraceutical value. Accurate yield prediction is vital for optimizing cultivation practices and market planning. This study employs three regression models, i.e. linear regression, decision tree regression and random forest regression, on a dataset that includes tree height (m), age of tree, canopy volume (m3), rootstock and scion girth (m), total number of fruits per tree, average fruit weight (g) and yield per plant (kg) to predict the yield of loquat trees under subtropical conditions of Punjab state. Data used for model training and validation were recorded during 3 consecutive years (2023–2025) from loquat trees planted at 6.5 × 6.5 m spacing in the college orchard at Punjab Agricultural University, Ludhiana. It was concluded that, based on coefficient of determination, root mean square error and mean absolute error values of model evaluation metrics, the decision tree regression model demonstrates high predictive accuracy among three models, offering a valuable tool for growers and researchers to predict loquat yield. The random forest regressor performed equally well, slightly underperforming compared to the decision tree model. These models suggests that machine learning approaches are robust tools for orchard management and precision agriculture in loquat cultivation.