I have a data set consisting of a csv-file with three columns with numerical data.
I have performed my own implementation that clusters the data set with K-means and then calculates outliers based on euclidean distance between data points and the cluster centroids.
I wan't to perform the same kind of operation in Splunk but have not been successfull so far.
I have tried local outlier factor, with the following query in search:
source="dataset.csv" | fit LocalOutlierFactor 0,1,2 | search isOutlier="1.0"
However, the result from this search is very poor since very few outliers are detected. The data set is labeled making it easy to see correctly classified outliers.
I have also tried with "Detect numeric outliers" from the machine learning toolkit but there, I can only chose one field to analyze and I have three fields.
Is there an optimal solution to the problem of finding outliers in this type of dataset?
Thanks in advance!