Classification and discriminant analysis of maize heterosis groups: A methodological system based on moderate SNP markers and machine learning

Rational utilization of maize heterosis is a core pathway to enhance yield potential and ensure food security, and the accuracy of heterosis group classification is directly related to the scientific prediction of heterosis and parental selection. Traditional approaches are limited by the number of molecular markers and statistical models, making it difficult to balance accuracy and stability. In this study, waxy maize inbred lines were used as materials, and marker sets with different densities were constructed based on high-throughput SNP markers. Population structure analysis and genetic clustering were applied for group classification, while random forest and support vector machine algorithms were introduced for discriminant analysis and cross-validation. The results showed that moderate-density markers outperformed high-density markers in terms of within-group consistency and clustering stability; in discriminant analysis, the random forest model achieved the highest prediction accuracy, exceeding that of the support vector machine. These findings indicate that excessively high marker density does not necessarily improve classification performance, and that moderate marker density combined with machine learning methods enables more efficient and stable group classification and prediction. This study established a methodological system for maize heterosis group classification and discriminant analysis based on moderate SNP markers and the random forest algorithm, providing new technical support for rational parental selection and heterosis prediction in maize breeding.