Prediction of greenhouse tomato yield using artificial neural networks combined with sensitivity analysis

In this paper, artificial neural networks (ANNs) combined with sensitivity analysis was applied to predict greenhouse tomato yield (Lycopersicon esculentum Mill.) and bring out the most influencing inputs on tomato production. The models were created using data randomly collected by a face-to-face survey from 25 greenhouse tomato farms in Biskra Province, Algeria. Results showed that total energy inputs were 94 748.2 MJ.ha−1 while energy ratio was 1.055 indicating a relatively low energy efficiency. Different ANN models were tested by varying the number of neurons in the hidden layer from 1 to 40. Based on statistics criteria, the best structure found is 12–34–1. This ANN model was used to estimate tomato yield using the energy inputs. Then, the results of ANN model were compared with those from the multiple linear regression (MLR) technique. The results illustrated that ANN provided more accurate predictions than the MLR technique. Sensitivity analysis revealed that insecticides, farmyard manure (FYM), potassium (K2O), nitrogen (N), electricity and fungicides are the most significant inputs in the greenhouse tomato production. In conclusion, ANNs are simple and effective tools for predicting tomato yield and extracting knowledge via sensitivity analysis.