Abstract
Abstract In this paper, we propose an Artificial Neural Network (ANN)-based model that can be used to predict the natural rubber concentration in rubber dandelion ( Taraxacum kok-saghyz) roots. Morphological characteristics of the plants serve as the input for the neural network, whereas the predicted variable is the rubber yield. Analysis of plant accessions HOTEL and OH1004 indicated that eight of the investigated physiological traits correlated closely with the targeted rubber concentration. The architecture of the model was tailored to this input data and subsequently trained on measured values. Post-training network analysis was also performed, which helped identify the morphological characteristics that influenced rubber concentration the most. To broaden the application areas of the model, a generalized architecture is also proposed, which leverages the flexibility of neural networks and simultaneously uses quantitative and qualitative data as input. The measured vs. calculated rubber concentration correlation plots have shown that the generalized model is capable of providing reliable results for multiple plant accessions.