Relative Density-based Outlier Factor (RDOS) algorithm with gaussian kernel
Function to calculate the Relative Density-based Outlier Factor (RDOS) as an outlier score for observations. Suggested by Tang, B. & Haibo, He. (2017)
RDOS(dataset, k = 5, h = 1)
dataset |
The dataset for which observations have an RDOS score returned |
k |
The number of k-nearest neighbors used to identify reverse- and shared nearest neighbors |
h |
Bandwidth parameter for gaussian kernel. A small h put more weight on outlying observations |
RDOS computes a kernel density estimation by combining the nearest, reverse nearest and shared neighbors into one neighborhood. The density estimation is compared to the density estimation of the neighborhoods observations. A gaussian kernel is used for density estimation, given a bandwidth chosen by user. A kd-tree is used for kNN computation, using the kNN() function from the 'dbscan' package.
It is a computational heavy task to identify reverse and shared neighbors from the kd-tree. Thus, the RDOS has high complexity and is not recommended to apply to datasets with n>5000. The RDOS function is useful for outlier detection in clustering and other multidimensional domains
A vector of RDOS scores for observations. The greater the RDOS score, the greater outlierness
Jacob H. Madsen
Tang, B. & Haibo, He. (2017). A local density-based approach for outlier detection. Neurocomputing. pp. 171-180. DOI: 10.1016/j.neucom.2017.02.039
# Create dataset X <- iris[,1:4] # Find outliers by setting an optional k outlier_score <- RDOS(dataset=X, k=10, h=2) # Sort and find index for most outlying observations names(outlier_score) <- 1:nrow(X) sort(outlier_score, decreasing = TRUE) # Inspect the distribution of outlier scores hist(outlier_score)
Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.