Enhancing Thermal Property Prediction of Corn Kernels: Integrating Cultivar Variability and Machine Learning Models

Abstract

Advancements in plant breeding and genetic engineering have led to increased productivity. However, limited research has assessed whether the thermal properties of these novel cultivars align with established empirical predictive literature. This study evaluated the thermal conductivity, specific heat, and thermal diffusivity of ten (10) U.S. grown corn cultivars across five moisture content levels (13% to 21% wet basis). The study further examined the predictive capability of empirical moisture and proximate composition-based models and machine learning models for estimating thermal properties of corn cultivars. The results showed a consistent increase in thermal properties with rising moisture content across all evaluated corn cultivars. Analyses revealed that both moisture content and cultivar significantly influenced thermal properties, confirming the complex and cultivar-specific nature of thermal behavior in corn. While models based on moisture content and proximate composition provided descriptive value, they exhibited limited predictive accuracy and generalizability, particularly across diverse cultivars. Notably, Gaussian Process Regression (GPR) models significantly outperformed Linear Regression (LR) models, with over 15% accuracy improvement in mean absolute error (MAE) and root mean square error (RMSE). Among the GPR approaches assessed, the Matern 5/2 kernel (GPR-52) achieved the highest accuracy and demonstrated robust, consistent performance, particularly in predicting thermal conductivity. The results emphasize the importance of integrating advanced modeling approaches and considering cultivar-specific and compositional variability in thermal processing applications. Future research should prioritize model refinement through the inclusion of additional explanatory variables such as bulk density and structural properties to further enhance prediction accuracy and practical applicability in agricultural and industrial settings.