MACHINE LEARNING FOR CROP SCIENCE: APPLICATIONS AND PERSPECTIVES IN MAIZE BREEDING

Abstract

Machine learning (ML) has been a major driver in complex data analysis in recent decades, allowing the mining of large databases. ML techniques allow the creation of computational models for prediction, pattern extraction and recognition, considering the premise that the computer acquires learning skills to perform a given task without being explicitly programmed for such a purpose. Driven by the efficiency of these techniques, several studies have demonstrated their wide range of applications and high potential for maize breeding. From the prediction of genetic values by omic data to applications of high-throughput phenotyping data, ML models have promoted advances in the species comprehension and assisted in the development of more effective tools for its breeding, driving expressive yield gains. In this context, this work presents the main contributions of ML in maize breeding, providing a broad view of the main studies and methodological perspectives in the area.