Characterization of Iranian Grapevine Cultivars Using Machine Learning Models

Rapid and precise identification and distinguishing of clones and cultivars are critical issues in grape (Vitis vinifera L.). In this study, the efficiency of supervised machine learning algorithms to classify and predict Iranian grapevine cultivars was evaluated. Seventy grape genotypes including five European grape varieties and 65 Iranian grape cultivars were included for SSR profiling. To the selection of the most demonstrative and discriminating features (alleles), sixteen SSR data were then subjected to five weighting algorithms. Then, induction tree and Naive Bayesian methods were employed for model construction. The results indicated that the induction tree and Naive Bayes models classify and predict the cultivars with high accuracy (> 80%). Moreover, graphical models of the decision tree and weighting algorithms introduced the VVMD7_1_236, VVMD5_2_238, and ZAG47_2_167 as more important discriminative alleles for Local/National and Iranian/International cultivars prediction, respectively. The present study provided details on allele diversity by identification of effective polymorphic alleles in cultivars characterization using the supervised machine learning methods. Moreover, inexpensive, rapid and effective analytical tools were developed for future grapevine conservation and germplasm management programs. © 2019, The National Academy of Sciences, India.