Combining linear regression and machine learning approaches to identify consensus variables related to optimum sweetpotato transplanting date

Abstract

Forward and stepwise regression methods identified variables related to the influence of transplanting date on yield of U.S. #1 sweetpotatoes. The variables were mean minimum soil temperature 5 days after transplanting (DAT), wind direction at transplanting, and accumulated heat units (growing degree-days) 5 DAT. Machine learning techniques identified the same variables using leave-one-out and k-fold cross-validation methods. Growers and crop consultants, in collaboration with knowledge workers, can use this information in conjunction with public and subscription-based weather forecasts to further optimize transplanting date determination and for making risk-averse decisions. These results help to underscore the importance of consistent transplant establishment as one of the determinants of storage root yield in sweetpotatoes.