One class random forests

Abstract

One class classification is a binary classification task for which only one class of samples is available for learning. In some preliminary works, we have proposed One Class Random Forests (OCRF), a method based on a random forest algorithm and an original outlier generation procedure that makes use of classifier ensemble randomization principles. In this paper, we propose an extensive study of the behavior of OCRF, that includes experiments on various UCI public datasets and comparison to reference one class algorithms -namely, gaussian density models, Parzen estimators, gaussian mixture models and One Class SVMs -with statistical significance. Our aim is to show that the randomization principles embedded in a random forest algorithm make the outlier generation process more efficient, and allow in particular to break the curse of dimensionality. One Class Random Forests are shown to perform well in comparison to other methods, and in particular to maintain stable performance in higher dimension, while the other algorithms may fail.

Can I Trust My One-Class Classification?

“…It is worth noting that classifier ensembles have also been applied successfully in the field of species distribution modeling [28,29]. Furthermore they are a focus of intense research in pattern recognition and machine learning [30,31]. These are important developments because multiple classifier systems have been shown to be successful supervised classification of remote sensing data [32][33][34] and should be further investigated for one-class classification.…”

Forming Ensembles of Soft One-Class Classifiers with Weighted Bagging

“…8,9,14,36) However, they do not take into consideration the specific nature of OCC problems which results in their varied performance. There are some works done on introducing pruning to OCC ensembles 7,21,29) and dedicated diversity measures (as the standard ones tend to fail in this task).…”

Pixel-Based Unsupervised Classification Approach for Information Detection on Optical Markup Recognition Sheet

“…This paper also used the five accuracy parameters [20], which were the Predictive Accuracy Rate (PACC), Recall/True Positive/Sensitivity Rate (RR), Specificity/True Negative Rate (SR), Precision/Positive Predictive Value (PPV), Negative Predictive Value (NPV). The Matthews' Correlation Coefficient (MCC) was also used [21]. They were calculated as follows: [21] and [22] whereby Figure 2: Classifiers to Measure Accuracy for Machine Binary Classification.…”

“…The Matthews' Correlation Coefficient (MCC) was also used [21]. They were calculated as follows: [21] and [22] whereby Figure 2: Classifiers to Measure Accuracy for Machine Binary Classification. [23] Conferring to these equations, [23] incorporated them into their contingency confusion table.…”

“…The confusion matrix was drawn using recordings on the number of circle bubbles that are capable of being detected by the proposed algorithm when tested under both phases. This confusion matrix, as modified by [24] with the six parameters of [20] and [21], is shown in Figure 8.…”