Discriminating among tectonic settings of spinel based on multiple machine learning algorithms

Figures & data

Figure 9. ROC and AUC of Bag.

Figure 10. ROC and AUC of RUSBoost.

Figure 11. ROC and AUC of Weighted SVM with Gaussian kernel.

Figure 1. Distribution of spinel samples.

Figure 2. Projection process of LDA.

Figure 3. Classification process of SVM.

Figure 4. Classification process of KNN.

Figure 5. Basic framework of an Ensemble Classifier.

Figure 6. Illustration of ROC and AUC.

Figure 7. The interface of the classification learner application of MATLAB.

Table 1. Optimal parameters and validation accuracies of different classifiers.

	Classifier	Optimal parameters	Validation accuracy
Decision Trees	Fine Tree	Maximum number of splits = 150	78.20%
	Medium Tree	Maximum number of splits = 150	78.30%
	Coarse Tree	Maximum number of splits = 150	78.30%
Discriminant Analysis	LDA	Covariance structure: Full	69.00%
	QDA	Covariance structure: Full	67.70%
SVMs	SVM (Gaussian kernel)	Box constraint level = 5	84.70%
	SVM (Linear kernel)	Box constraint level = 1	73.70%
	SVM (Quadratic kernel)	Box constraint level = 3	80.90%
	SVM (Cubic kernel)	Box constraint level = 4	82.90%
Nearest Neighbor Classifiers	Fine KNN	Number of neighbors = 3	82.00%
	Medium KNN	Number of neighbors = 3	81.00%
	Coarse KNN	Number of neighbors = 3	81.60%
	Cosine KNN	Number of neighbors = 3	81.50%
	Cubic KNN	Number of neighbors = 3	82.50%
	Weighted KNN	Number of neighbors = 7–13	84.4%
Ensemble Classifiers	AdaBoost	Maximum number of splits = 100; Number of learners = 800	86.60%
	Bag	Maximum number of splits = 150; Number of learners = 1000	86.30%
	RUSBoost	Maximum number of splits = 100; Number of learners = 800	85.00%
	Subspace (Decision Tree)	Subspace dimension = 5; Number of learners = 800	80.70%
	Subspace (Discriminant)	Subspace dimension = 7; Number of learners = 800	68.00%

Figure 8. ROC and AUC of AdaBoost.

Figure 12. ROC and AUC of Weighted KNN.

Table 2. AUCs of AdaBoost, Bag, RUSBoost, SVM with Gaussian kernel, and Weighted KNN.

Classifier	Validation accuracy	AUC (OI-CM & SC)	AUC (CM-OI & SC)	AUC (SC-OI & CM)	Average AUC
AdaBoost	86.60%	0.94	0.93	0.94	0.937
Bag	86.30%	0.96	0.94	0.97	0.957
RUSBoost	85.00%	0.95	0.94	0.96	0.950
SVM with Gaussian kernel	84.70%	0.94	0.92	0.96	0.940
Weighted KNN	84.40%	0.93	0.91	0.95	0.930
Average	85.12%	0.944	0.928	0.956	0.943

Figure 13. Confusion matrix of Bag Ensemble classifier and AdaBoost Ensemble classifier.

Figure 14. Evaluations of the importance of major elements.

Figure 15. Interface of DSTS.

Figure 16. Location of DSTS application.