Open Access

Table B.1.

Set of best values obtained from a grid search-based optimization of the five selected hyperparameters, for each RF classifier tested in this work.

RF classifier n_estimators min_samples_split min_sample_leaf max_depth max_features
g, spec-MIR LS 300 10 4 10 sqrt
r, spec-MIR LS 100 2 4 10 sqrt
i, spec-MIR LS 500 10 4 10 sqrt

g, main LS 300 10 4 10 sqrt
r, main LS 300 10 4 10 sqrt
i, main LS 100 10 4 10 sqrt

g, var. features only 300 2 4 10 log2
r, var. features only 500 10 2 30 sqrt
i, var. features only 100 10 2 10 log2

uBrizy-derived colors only 300 10 4 10 sqrt
all colors only 100 10 2 20 sqrt

g, 13 features 100 10 4 20 sqrt
r, 14 features 100 10 4 20 sqrt
i, 10 features 500 2 4 10 sqrt

Notes. The chosen hyperparameters to optimize (see Section 3) are: n_estimators (tested values: 100, 300, 500), min_samples_split (tested values: 2, 5, 10), min_samples_leaf (tested values: 1, 2, 4), max_depth (tested values: 10, 20, 30, None), and max_features (tested options: sqrt and log2). The first column lists the classifiers evaluated in this work, with their performance metrics shown in Table 3.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.