A Machine Learning and Cross-Validation Approach for the Discrimination of Vegetation Physiognomic Types Using Satellite Based Multispectral and Multitemporal Data

Table 4

Cross-validation results (size of ground truth data sets for each class = 300). The Standard Deviations (SD) across the 10-fold cross-validation in the case of optimum number of features are also shown.

Number of experiments	Max. overall accuracy	Overall accuracy SD	Max. kappa coefficient	Kappa coefficient SD	Optimum number of features
1	0.79	0.03	0.75	0.04	92
2	0.79	0.03	0.75	0.04	99
3	0.70	0.03	0.64	0.04	106
4	0.80	0.03	0.76	0.03	153
5	0.81	0.03	0.77	0.03	98
6	0.81	0.03	0.78	0.03	160
7	0.80	0.03	0.76	0.04	210
8	0.79	0.04	0.75	0.05	211
9	0.80	0.03	0.76	0.04	211
10	0.80	0.03	0.76	0.04	134