Machine learning applied to electrical conductivity data quickly predicts percentage germination of soybean seed lots
Charline Zaratin Alves, Amanda Maria Frata Ferreira, Alan Mário Zuffo, Salvador Barros Torres, Márcio Dias Pereira, Josué Bispo da Silva · Seed Science and Technology · 2025
Rapid vigour tests combined with machine learning (ML) techniques can provide quick and accurate results, optimising decision-making processes regarding seed lot management. The objective of this study was to analyse the performance of ML algorithms combined with rapid vigour tests in predicting soybean seed germination. The algorithms used were linear regression (LR), artificial neural network (ANN), support vector machine (SVM), k‐nearest neighbor (KNN), M5P, random forest (RF), RandTree and REPTree. As input, electrical conductivity and tetrazolium tests (vigour ‐ TVIG and viability ‐ TVIA) were considered, alone and in combination. The correlation coefficient ( r ) and mean absolute error (MAE) metrics were used to evaluate the performance of the tested prediction models. The ANN and M5P algorithms achieved the highest r values (0.86). For the inputs, the highest r values (0.89) were for electrical conductivity, electrical conductivity + TVIA, electrical conductivity + TVIG and electrical conductivity + TVIA + TVIG. The M5P model had lower MAE values (5.05), followed by RF (5.31) and REPTree (5.60) for the electrical conductivity, electrical conductivity + TVIG and electrical conductivity + TVIA inputs. The electrical conductivity test combined with the M5P algorithm is an excellent alternative for the rapid prediction of soybean seed germination.