Machine Learning for Non-Destructive Prediction of Sunflower Leaf Area

Leaf area is an important parameter in plant growth, physiology, and productive potential studies. However, its measurement using traditional methods can be limited. The search for non-destructive approaches based on leaf dimensions is essential, and machine learning offers promising alternatives for accurate estimation. Therefore, the present study aimed to develop and compare linear regression models and machine learning algorithms for the non-destructive prediction of leaf area in four sunflower cultivars (BRS 323, Altis 99, Sany 66, and BRS 442). The best-performing model was also made available through an interactive and freely accessible web interface. Models based on artificial neural networks (ANN), random forest, and linear regression were developed using leaf length and width as input variables, and leaf area as the output. Model performance was evaluated using the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and Willmott’s agreement index (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">d</i>). The ANN model with a 2-4-1 architecture showed the highest accuracy (R² = 0.9937; RMSE = 1.1060; MAE = 0.7851; MAPE = 4.7916; <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">d</i> = 0.9984), outperforming the other models. This ANN model was subsequently implemented in the web interface, offering an efficient, non-destructive tool for estimating sunflower leaf area based on linear dimensions.