Machine Learning Applications on Agricultural Datasets for Smart Farm Enhancement

Method: NN 
Apple 
Pears 
Italian Province Real Value Predicted Value Real Value Predicted Value 
L’Aquila 45,900 45,000 3925 
3750 
3.2. Task 2—Comparison between Machine Learning Algorithms on Missing Data (CNR Dataset—Results) 
For this task, the predictive errors depicted in Table 6 highlight that the polynomial model best 
fits the LAI values prediction for the three considered cultures. This outcome can be explained 
considering the nature of this scientific values, as well as the temporal discontinuity with which they 
have been gathered, along with their small amount; for the polynomial model, there is a very large 
difference from the others, highlighting the simplicity and the advantage of using this standard but 
also the performing technique. 
The plot comparison between linear and polynomial predictive models on this scientific dataset 
is in Figure 7, where a polynomial interpolation (green plot) shows how the predictive model is able 
to approximate the peculiar growing trend (blue plot), which can fit unknown incoming data very 
well. The higher grade mathematical model is better than the others and this happens both if, for the 
training, you give the data for a single year, either if you give data for three years. 
Table 6. 
Task 2: comparison on the prediction error for the cultures leaf area index (LAI) value 
exploiting machine learning methods on the CNR scientific agrarian dataset. 
Culture
Prediction Error 
NN LR 
Polynomial 
Artichokes 139.00% 
101.63% 
25.70% 
Pear 1779.38% 
81.80% 
10.00% 
Pacciamata Eggplant 933.10% 564.89% 
6.26% 
Figure 7.
Task 2: plot comparison between real-values (red dots), the linear (blue), and polynomial 
(green) predictive model on the CNR scientific agrarian dataset.