EcoSta 2022: Start Registration
View Submission - EcoSta2022
A0891
Title: Prediction of flowering and maturity time of soybean using stacking Authors:  Akio Onogi - Ryukoku University (Japan) [presenting]
Abstract: In crop breeding, it is important to control phenological traits according to target regions because phenological traits are related to local adaptation. Typical phenological traits, flowering and maturity times, are known to be affected by genetic and environmental factors such as temperature. To predict the flowering and maturity time of new cultivars accurately, an ensemble learning, stacking, was applied to soybean data and compared with other methods. The explanatory variables included daily mean, maximum, and minimum temperature, precipitation, hours of sunshine, and day length as environmental factors, and genotypes of five genes relevant to flowering. The response variables were days from sowing to flowering for flowering time, and days from flowering to maturity for maturity time. A total of 41 learners including random forests, cubist, gradient boosting, support vector machine, and elastic net were used as base models of stacking. Random forests and linear regression were compared as the meta-model. Besides stacking, each base model and method based on an eco-physiological model of crop phenology were also compared. The evaluation using independent data shows the superiority of stacking that used random forests as the meta-model among the methods compared.