Affordable Bimodal Optical Sensors to Spread the Use of Automated Insect Monitoring

Table 2

B. oleae versus C. capitata recognition results.

Classifiers	Scatter	Extinct	Both^#
	% mean acc./SD	% mean acc./SD	% mean acc./SD
Linear SVC^1	96.63/0.62	87.33/1.12	98.54/0.44
RBF SVM^2	94.73/0.42	84.02/1.42	97.68/0.47
XGBC	98.49/0.24	92.33/0.70	98.99/0.42
RF^3	97.53/0.60	91.51/0.85	98.02/0.52
X-TREE^4	97.35/0.48	91.14/0.74	98.28/0.47
GBC^5	98.41/0.35	92.33/0.76	98.86/0.37

¹Linear kernel, C = 0.01. ²Radian basis function kernel, gamma = 0.009, C = 0.2. ^3,4Number of trees = 650, min_samples_split = 2, min_samples_leaf = 1. ⁵min_samples_split = 5, min_samples_leaf = 30, max_depth = 4. ^#“Both” means the two PSD feature vectors are appended in column-wise order. Mean accuracy of top-tier classifiers using a 10-fold cross-validation scheme with 20% of the corpus was randomly holdout. Verification results are based on the dataset in Table 1. Note that each species contains both sexes. Mean and standard deviation of accuracy measure over all folds (% mean/SD over). Linear SVC: linear support vector classifier; RBF SVM: radial basis function support vector machine; XGBC: extreme gradient boosting classifier; RF: random forests; X-TREE: extra randomized trees; GBC: gradient boosting classifier.