Journal ID : AMA-17-11-2021-10846
[This article belongs to Volume - 52, Issue - 03]
Total View : 435

Title : Neural network analysis of mechanization level-a Case Study in Northwest regions of Tunisia

Abstract :

Agricultural mechanization level in two governorates in northwestern Tunisia has been assessed and analyzed. Interviews, observations, and a structured questionnaire were used for the collection of the database. Five artificial neural networks (ANN) models with two hidden layers were used to estimate the mechanization level. Initially, 80 attributes were used as input variables to predict the level of mechanization. The Forward regression method (SPSS 22 software) followed by the significance analysis method was used to select the typical variables. Ten variables were selected as models’ inputs. The performance of the ANN model was evaluated with various statistical measures including coefficient of determination (R2), mean square error (MSE), and mean absolute error (MAE). The optimal ANN models had correlations of 0.94 with calculated mechanization levels. Sensitivity analysis of the models showed that farm area, labor, dominant crop area, and the number of tractors is the typical inputs affecting the level of mechanization. The results presented in this study justify the low level of mechanization according to farm area (Maximum value less than 30%), dominant crop area (average value less than 25%), number of tractors (Maximum value less than 30%). However, the availability of labor gives an acceptable level (on average 50%) for farms with some skilled labor between 3 and 4 and a low mechanization level for farms with some skilled labor is less than 2 (on average 10%).

Full article