The goal of RobustEM is to carry out clustering on high dimensional points using expectation-maximization algorithm which is robust against outliers. To install the package, make sure you have devtools package loaded and type install_github("mkbwang/RobustEM", ref="main").

library(RobustEM)

The example here simulate 720 points belonging to 6 clusters. Each cluster has 120 points. All the points have two dimensions. All the clusters have approximately 6% of points as outliers. The outliers are generated here as having the same mean but a covariance matrix with much larger elements. This customized function simMultGauss not only generates the data points but also includes detailed cluster information (mean and covariance of each cluster).

set.seed(22)
sim_info <- simMultGauss(n = 120, d = 2, cluster = 6, out_perc = 0.03, out_mag = 4)

We can use function robustEM to cluster the points.

result <- robustEM(sim_info[["simdata"]], cluster = 6)

I have written a customized summary function to summarize the cluster mean and the number of points in each cluster.

summary(result)
#> $`Cluster Point Count`
#> point_cluster
#>   1   2   3   4   5   6 
#> 121 119 119 121 120 120 
#> 
#> $`Cluster Mean`
#>            V1       V2
#> [1,] 16.49681 24.26420
#> [2,] 18.45567 14.19157
#> [3,] 35.74641 12.52521
#> [4,] 30.25588 26.08767
#> [5,] 27.99846 41.28102
#> [6,] 48.01971 34.62843

plot(result)