Artificial neural network potential in yield prediction of lentil (Lens culinaris L.) influenced by weed interference

This study was conducted to predict the yield and biomass of lentil (Lens culinaris L.) af- fected by weeds using artificial neural network and multiple regression models. Systematic sampling was done at 184 sampling points at the 8-leaf to early-flowering and at lentil maturity. The weed density and height as well as canopy cover of the weeds and lentil were measured in the first sampling stage. In addition, weed species richness, diversity and even- ness were calculated. The measured variables in the first sampling stage were considered as predictive variables. In the second sampling stage, lentil yield and biomass dry weight were recorded at the same sampling points as the first sampling stage. The lentil yield and biomass were considered as dependent variables. The model input data included the total raw and standardized variables of the first sampling stage, as well as the raw and stan- dardized variables with a significant relationship to the lentil yield and biomass extracted from stepwise regression and correlation methods. The results showed that neural network prediction accuracy was significantly more than multiple regression. The best network in predicting yield of lentil was the principal component analysis network (PCA), made from total standardized data, with a correlation coefficient of 80% and normalized root mean square error of 5.85%. These values in the best network (a PCA neural network made from standardized data with significant relationship to lentil biomass) were 79% and 11.36% for lentil biomass prediction, respectively. Our results generally showed that the neural net- work approach could be used effectively in lentil yield prediction under weed interference conditions.