Comprehensive prediction of bread quality in hard red winter wheat using machine learning: integrating physicochemical and rheological properties

Authors: OLAGUNJU, OLUSOLA - Kansas State University; DU, ZHENJIAO - Kansas State University; Tilley, Michael; Wu, Xiaorong; CHEN, RICHARD - Retired ARS Employee; XU, XUAN - Kansas State University; ZHANG, GUORONG - Kansas State University; LI, YONGHUI - Kansas State University

Submitted to: Journal of Cereal Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/10/2026
Publication Date: 4/13/2026
Citation: Olagunju, O., Du, Z., Tilley, M., Wu, X., Chen, R., Xu, X., Zhang, G., Li, Y. 2026. Comprehensive prediction of bread quality in hard red winter wheat using machine learning: integrating physicochemical and rheological properties. Journal of Cereal Science. 129(104450). https://doi.org/10.1016/j.jcs.2026.104450.
DOI: https://doi.org/10.1016/j.jcs.2026.104450

Interpretive Summary: The aim of this research was to assess how ten different machine learning algorithms performed in predicting important bread quality features such as loaf volume, crumb texture, grain characteristics, and overall baking quality, using data from 359 cultivars of hard winter wheat (HWW). To achieve this, we assessed the models' accuracy with two distinct feature sets: (a) physicochemical properties of wheat flour alone; and (b) a combination of physicochemical and rheological properties. The goal of this comparative analysis was to identify the most effective feature set and machine learning model for predicting bread quality in HWW cultivars. The Random Forest model showed its strength by consistently outperforming other algorithms, highlighting the value of using both feature sets. Results show that predicting HWW wheat baking quality relies on a delicate balance between physicochemical and rheological factors, with sedimentation volume and extensograph energy as the top predictors of bread quality attributes. This information offers useful applications in wheat breeding projects and evaluations of wheat baking quality in the industry, as utilization of machine learning models and picking the right wheat properties can ease the wheat cultivars selection process.

Technical Abstract: Baking quality tests are a crucial screening tool in wheat breeding, ensuring that experimental wheat lines meet the end-use characteristics of their target market class. This study aimed to evaluate the capability of machine learning models to predict key wheat-baking quality traits, such as bread loaf volume, crumb grain, crumb texture, and overall baking quality, using a dataset of 359 Hard Red Winter (HRW) wheat varieties. Two feature sets were employed: a base set containing only physicochemical properties and an alternative set integrating both physicochemical and rheological properties. With only physicochemical properties, the support vector machine model achieved a test accuracy of 85% in predicting bread volume into three classes: low (663–942 cc), moderate (943–1080 cc), and high (1081–1325 cc). Incorporating rheological properties improved the prediction of overall baking quality into ‘good and excellent’ quality and achieved a test accuracy of 93%. The overall baking quality encompasses various aspects of bread quality, including bread loaf volume, shape and crumb characteristics. While rheological properties added complexity to the models, they were particularly impactful in predicting other attributes like crumb grain and texture. Feature importance analysis revealed sedimentation volume and extensograph energy as the top predictors of bread quality attributes. This study emphasizes the necessity of leveraging multiple parameters, as no single feature can reliably predict HRW wheat baking quality. The models developed here offer an effective tool for screening wheat experimental lines during the final stages of variety development.