Discrimination of onion subjected to drought and normal watering mode based on fluorescence spectroscopic data

•The fluorescence spectroscopic data proved to be useful for the discrimination of onion samples.•Samples growing under drought and normal watering were discriminated with an accuracy reaching 100%.•Different onion varieties and lines were correctly distinguished in 90% for samples subjected to drought.•The accuracy reached 84% for discrimination of varieties and lines growing under normal watering. Drought stress can affect the yield and quality of cultivated plants. The deficit of water may result in the physiological and anatomical reactions at organ, tissue and cellular levels of the plant species. The objective of this study was to discriminate different onion samples with the use of innovative models based on fluorescence spectroscopic data using different classifiers. The onion growing under drought and normal watering conditions were compared. Additionally, the five different samples of onion including three varieties (Konkurent bql, Asenovgradska kaba, Trimoncium) and two lines (white, red) subjected to both the drought mode and normal watering mode were differentiated. The results were evaluated based on confusion matrices, average accuracies, and the values of TP (True Positive) Rate, FP (False Positive) Rate, Precision, F-Measure, ROC (Receiver Operating Characteristic) Area and PRC (Precision-Recall) Area. In the case of the discrimination of two classes: drought mode and normal watering mode, an average accuracy reached 100% for white line of onion for a model built using the Naive Bayes, Multilayer Perceptron, JRip and LMT classifiers and for red line of onion for all used classifiers (Naive Bayes, Multilayer Perceptron, IBk, Multi Class Classifier, JRip, LMT). The values of TP Rate, Precision, F-Measure, ROC Area and PRC Area were equal to 1.000, and FP Rate was 0.000. For onion samples subjected to drought, five classes including the Konkurent, Asenovgradska kaba, Trimoncium varieties and the white and red lines were discriminated with an average accuracy of up to 90% for the LMT classifier. The same classes of samples but subjected to normal watering were correctly distinguished in 84% for the Naive Bayes classifier.