The Application of Multiple Linear Regression and Artificial Neural Network Models for Yield Prediction of Very Early Potato Cultivars before Harvest

Yield forecasting is a rational and scientific way of predicting future occurrences in agriculture—the level production effects. Its main purpose reducing the risk decision-making process affecting yield terms quantity quality. The aim following study was to generate linear non-linear model forecast tuber three very early potato cultivars: Arielle, Riviera, Viviana. In order achieve set goal study, data from period 2010–2017 were collected, coming official varietal experiments carried out northern northwestern Poland. has been created based on multiple regression analysis (MLR), while built using artificial neural networks (ANN). models can predict varieties 20th June. Agronomic, phytophenological, meteorological used prepare models, correctness their operation verified basis separate sets not participating construction models. For proper validation model, six error metrics used: i.e., global relative approximation (RAE), root mean square (RMS), absolute (MAE), percentage (MAPE). As result conducted analyses, results for most did exceed 15% MAPE. predictive NY1 characterized by better values quality measures ex post errors than RY1.