Hazelnut (Corylus avellana L.) Mass Prediction Using Machine Learning Models Based on Physical Attributes

Türkiye is the dominant global producer of hazelnuts, supplying approximately 70% of the world’s total output. The physical characteristics of hazelnuts, particularly mass, play a critical role in determining quality and shelf life. To improve efficiency in hazelnut handling, this study explores non-destructive, rapid prediction of hazelnut mass using machine learning (ML) models based on physical attributes such as size and shape. A total of 1,050 hazelnuts from seven cultivars (Çakıldak, Kalınkara, Palaz, Sivri, Tombul, Yerli, and Yomra), collected from orchards in Türkiye's Black Sea region, were analyzed. Each sample was measured for length, width, and thickness using digital calipers, and weighed with ±0.001 g precision. Derived attributes—including volume, geometric mean diameter, projected area, surface area, sphericity, shape index, aspect ratio, and elongation—were computed mathematically. Three ML models were evaluated: Linear Regression, Random Forest, and Artificial Neural Networks (ANN). Among these, the Artificial Neural Networks and Linear Regression models achieved the highest prediction performance (R² = 0.87, 0.871 for kernel, and 0.802, 0.804 for kernels, respectively). These findings highlight the potential of ML techniques in non destructive mass estimation of hazelnuts, with prospects for enhancement through the integration of chemical attributes for improved cultivar discrimination and mass prediction accuracy.