Non-destructive measurement of eggshell strength using NIR spectroscopy and explainable artificial intelligence

Eggshell strength is crucial for ensuring high-quality eggs, reducing breakage during handling, and meeting consumer expectations for freshness and integrity. Conventional methods of eggshell strength measurement are often destructive, time-consuming and unsuitable for large-scale applications. This study evaluated the potential of near-infrared (NIR) spectroscopy combined with explainable artificial intelligence (AI) as a rapid, non-destructive method for determining eggshell strength. Various multivariate analysis techniques were explored to enhance prediction accuracy, including spectral pre-processing and variable selection methods.

Principal component analysis and partial least squares discriminant analysis effectively classified eggs based on a threshold shell strength of 30 N. Regression models, including partial least squares regression, random forest (RF), light gradient boosting machine and K-nearest neighbors, were evaluated. Using only 14 selected variables, the RF model achieved a very good prediction performance with $R_p^2$ of 0.83, root mean square error of prediction of 1.49 N and ratio of prediction to deviation of 2.44. The Shapley additive explanation approach provided insights into variable contributions, enhancing the model's interpretability.

This study demonstrated that NIR spectroscopy, integrated with explainable AI, is a robust, non-destructive and environmentally sustainable approach for eggshell strength prediction. This innovative method holds significant potential for optimizing resource utilization and enhancing quality control in the egg industry. © 2025 The Author(s). Journal of the Science of Food and Agriculture published by John Wiley & Sons Ltd on behalf of Society of Chemical Industry.