A comparative evaluation of Bayes, functions, trees, meta, rules and lazy machine learning algorithms for the discrimination of different breeding lines and varieties of potato based on spectroscopic data

The objective of this study was to compare the usefulness of machine learning algorithms for distinguishing the potato lines and varieties based on selected fluorescence spectroscopic data. The potato tubers belonging to two breeding lines S 617 and S 716 and two varieties Trezor and Sante were examined. The discrimination analysis was performed using machine learning algorithms from different groups. The average accuracies, confusion matrices, and the F-Measure, Precision, PRC (Precision-Recall) Area, ROC (Receiver Operating Characteristic) Area and MCC (Matthews Correlation Coefficient) values obtained for models built using different algorithms were compared. The breeding lines and varieties of potato were discriminated with very high average accuracies equal up to 95% for the SMO (Sequential Minimal Optimization) algorithms (group of Functions), Naive Bayes (group of Bayes), Hoeffding Tree (group of Trees), Multi Class Classifier (group of Meta), PART (group of Rules), IBk (Instance-Based Learning with parameter k) (group of Lazy). Models developed with the use of selected algorithms allowed for distinguishing some potato lines and varieties with an accuracy of up to 100% and the values of the F-Measure, Precision, PRC Area, ROC Area and MCC reaching 1.000.