Sex determination of breeding bombyx mori L. silkworm lines based on different pupae, cocoon and silkmouth characteristics by using data mining algorithms

Sericulture is a very important branch of animal husbandry because it is easy and short-lived to raise silkworms, has high profitability, and produces a natural fiber source. This study aims to determine the sex of silkworms at an early stage by using Chi-Square Automatic Interaction Detector (CHAID), Exhaustive CHAID (Ex-CHAID), Classification and Regression Tree (CART) and Quick Unbiased, Efficient Statistical Tree (QUEST) tree-based data mining algorithms by using the silkmoth weight, some cocoon and pupa characteristics. Morphometric features of pupa length (PU), pupa weight (PA), pupa width (PG), cocoon width (KG), cocoon length (KU), cocoon shell weight (KGA), fresh cocoon weight (YKA), silkmoth weight (KA) of 400 male and female M and N pure line silkworms were used as the material of the study. In the statistical model, morphometric features were taken as independent variables, while gender was taken as dependent variable. The data set was divided into 70% training and 30% test set and tree-based data mining algorithms CHAID, Ex-CHAID, CART and QUEST were used. The classification performances of CHAID, Ex-CHAID, CART and QUEST algorithms used in determining the sexes by using pupal features were determined as follows; Area under the ROC curve (AUC±SE) was 0.800±0.0262, 0.833±0.0235, 0.814±0.0251, 0.796±0.0266, respectively, and sex determination success was 80.60%, 87.70%, 81.10%, 74.10%, respectively. In determining the sex of cocoon traits using tree-based data mining algorithms, it was determined that (AUC±SE) for the test set was 0.856±0.023, 0.808±0.015, 0.840±0.024, 0.864±0.022 and the success percentages in sex identification were 85.5%, 86.60%, 86.80%, 87.80%, respectively. The model fit criteria (AUC±SE) for the test set of algorithms for sex determination of pupa, cocoon and silkmoth (all) features were determined as 0.909±0.0188, 0.875±0.0211, 0.902±0.0186, 0.868±0.0221, and the performance results of the algorithms were determined as 86.20%, 88.10%, 83.30%, 89.10%, respectively. As a result; it was determined that the gender prediction performance by using pupa, cocoon and all features were Ex-CHAID, QUEST and CHAID algorithms, respectively.