Predicting soybean seed germination using the tetrazolium test and computer intelligence

Abstract

Seed quality is critical to agricultural productivity, and traditional testing can be time-consuming and subjective. Therefore, the use of machine learning can provide an efficient approach for predicting germination. The aim of this work was to investigate algorithms that, together with tetrazolium test data, lead to efficient prediction of soybean seed germination. The experiment was based on the collection and transcription of a database of thousand soybean seed analysis samples containing information on germination and tetrazolium tests (vigor and viability). The algorithms tested were REPTree, M5P, random forest, logistic regression, artificial neural networks and support vector machine, and the inputs tested were viability, vigor and vigor + viability (tetrazolium test) data. The data analysis used the correlation coefficient and mean absolute error as accuracy parameters of the algorithms. The results highlighted the support vector machine as the most effective algorithm for predicting germination, with the viability and vigor + viability inputs showing the best results. This study suggests that the integration of computational intelligence techniques with the tetrazolium test can make the assessment of soybean seed quality more efficient and contribute to fast and efficient decision making in agriculture.