Table 1.
Dimension Reduction Classifier Performance Summary.
Figure 1.
Misclassification rates of dimension reduction classifiers using the untrimmed datasets.
Mean misclassification rates for each of the dimension reduction-based methods using the full dataset (all variables) in the dataset to build the classification model. A) Is from the OC dataset [16], B) is from the Gaucher disease dataset [46], C) is from the LC datasets and D) is from the CRC dataset [14]. Blue circles illustrate PLS-LDA classification results, red triangles are from a PLS-RF classifier and purple crosses show results obtained from a PCA-LDA classifier.
Figure 2.
Misclassification rates of dimension reduction classifiers using the trimmed datasets.
Mean misclassification rates for each of the dimension reduction-based methods using the trimmed dataset to build the classification model. A) Is from the OC dataset [16], B) is from the Gaucher disease dataset [46], C) is from the LC datasets and D) is from the CRC dataset [14]. Blue circles illustrate PLS-LDA classification results, red triangles are from a PLS-RF classifier and purple crosses show results obtained from a PCA-LDA classifier.
Table 2.
SVM tuning results.
Figure 3.
Summary of Misclassification results for all classifiers.
Summary of mean MCR results for each of the optimised classifiers on each trimmed dataset. These results demonstrate the MCR for each classifier using the optimal number of reduced components from the PLS-LDA (excluding SVM). Gaucher data uses a 7 component model, the LC data uses a 5 component model, the CRC data uses a 6 component model and the OC data uses a 3 component model for each.
Figure 4.
Comparison of PLS and PCA for dimension reduction.
These plots demonstrate the capacity PLS has to separate classes based on the top 30 variables (Figure 4A) in the Gaucher dataset when compared to PCA (Note that this class separation is being heavily influenced by the loadings highlighted in Blue. Additionally, the vectors highlighted in red explain the within class variation in the control group. This is a key advantage PLS has over other methods.