Machine learning approaches to classify melon landraces based on phenotypic traits

Abstract

Phenotyping of native cultivars is becoming more essential, as they are an important for breeders as a genetic source for breeding. The variability of morphological properties plays critical role in melon breeding. In this paper various machine learning approaches were implemented to identify melon accession classes. A field experiment was conducted in Zahak Agriculture station to differentiate 144 melon accessions based on 14 traits. For this, Partial Least Square Discriminant Analysis (PLS-DA), Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbor (KNN) and Classification And Regression Trees (CART) were compared. The most commonly used performance values comprise overall accuracy, kappa value, Receiver Operating Characteristics (ROC) and Area Under Curve (AUC) were performed to identify accuracy of the models. The results showed the best performance for CART than others. The AUC and kappa value were 0.85 and 0.80 and fruit weight was the most important trait that affecting diversity in melon accessions. Regarding to these results Classification And Regression Trees (CART) is reliable for identification of melon accessions classes.