TrainSel: An R Package for Selection of Training Populations

Area of Science:

• Machine Learning
• Bioinformatics
• Quantitative Genetics

Background:

• Supervised learning applications are expanding, but face a major challenge due to insufficient and unrepresentative labeled training data.
• This data limitation is particularly prominent in fields like genomic selection and plant breeding, impacting prediction model performance.
• Careful selection of training data is crucial for enhancing the accuracy of predictive learning tasks.

Purpose of the Study:

• To introduce TrainSel, an R package designed for the selection of training populations (STP).
• To provide flexible, efficient, and user-friendly tools for optimizing training data selection.
• To demonstrate the utility and performance of TrainSel across various supervised learning applications.

Main Methods:

• Development of the TrainSel R package, offering specialized functions for STP.
• Application and evaluation of TrainSel in four distinct supervised learning scenarios.
• Comparative analysis of prediction accuracies using different training data selection strategies.

Main Results:

• TrainSel facilitates efficient and effective selection of training populations.
• The package demonstrates improved prediction accuracies in the tested applications.
• Performance was validated in both plant breeding and other relevant domains.

Conclusions:

• The TrainSel package addresses the critical need for optimized training data selection in supervised learning.
• It offers a practical solution for enhancing model performance where labeled data is scarce.
• TrainSel shows significant potential for advancing applications such as genomic selection and beyond.