Predicting sweetpotato traits using machine learning: Impact of environmental and agronomic factors on shape and size

• DOI: 10.1016/j.compag.2024.109215
• Publication Date: 2024-07-24T16:42:51Z
• Authors: Hangjin Liu; Shelly Hunt; G. Craig Yencho; Kenneth V. Pecota; Russell Mierop; Cranos M. Williams; Daniela S. Jones
• Keywords: Predictive modelling

Consumer preference in produce, defined by shape and size, heavily influences this market. Understanding the environmental management factors that impact these features can improve a farmer's economic margins. Since sweetpotatoes are hand-harvested tend to have varying shapes sizes, result unpredictable profit Methods for predicting aesthetic characteristics of using agronomic with machine learning not been developed. Moreover, crop size agricultural data analysis is challenging due need integrating diverse complex datasets, including genotypes, weather, field management, spatial information, into predictive models. This study employed an iterative process involving preparation, feature engineering, variable selection, model selection develop models predict sweetpotato traits from inputs. We collected organized various sources different formats, spatial, temporal resolutions. After comparing performance methods cross validation, Bagging regression had least error terms RMSE MAE sweetpotato's length-to-width ratio (RMSE = 0.185, 0.147) curvature 0.013, 0.010) predictions. outperformed naive baseline 29%–38% when Our also determined Covington cultivar GPS locations were most important influenced sweetpotatoes. Fertilizer prior planting, rose as curvature. Precipitation greater on prediction compared The methodology presented herein could be applied other crops like cucumbers, eggplants, peppers, potatoes, where determining their value.