Predicting yield is essential for producers, stakeholders and international interchange demand. The majority of the divergence in yield and essential oil content is associated with environmental aspects, including weather conditions, soil variety and cultivation techniques. Therefore, aniseed production was examined in this study. The categorical input variables for artificial neural network modelling were growing year (two successive growing years), growing locality (three different locations in Vojvodina Province, Serbia) and fertilization type (six different treatments). The output variables were morphological and quality parameters, with agricultural importance such as plant height, umbel diameter, number of umbels, number of seeds per umbel, 1000-seed weight, seed yield per plant, plant weight, harvest index, yield per ha, essential oil (EO) yield, germination energy, total germination, EO content, as well as the share of EOs compounds, including limonene, cis-dihydro carvone, methyl chavicol, carvone, cis-anethole, trans-anethole, β-elemene, α-himachalene, trans-β-farnesene, γ-himachalene, trans-muurola-4(14),5-diene, α-zingiberene, β-himachalene, β-bisabolene, trans-pseudoisoeugenyl 2-methylbutyrate and epoxy-pseudoisoeugenyl 2-methylbutyrate. The ANN model predicted agricultural parameters accurately, showing r2 values between 0.555 and 0.918, while r2 values for the forecasting of essential oil content were between 0.379 and 0.908. According to global sensitivity analysis, the fertilization type was a more influential variable to agricultural parameters, while the location site was more influential to essential oils content.
Keywords: aniseed; artificial neural network; essential oil; fertilization; growing year; locality.