-
This repo is the implementation of our methodology "One-Class Classification Ensembles with Unsupervised Representations to Detect Novelty". Outlier detection algorithms from the KDD field are employed to learn unsupervised representations and enrich the initial information that is enclosed in the feature space of a given dataset.
-
A benchmark data repository for outlier detection (http://www.dbs.ifi.lmu.de/research/outlier-evaluation/DAMI/) that is composed of 23 basic datasets was used to perform our experiments. The characteristics of the datasets is very diverse due to the fact that there is a large variety of applications in this data repository.
-
This repo uses the best of both worlds regarding the programming languages that were used. The R language was used to perform data manipulation and Python for the machine learning part. The integration between R & Python is made using the reticulate R library.
- First Run the src.R script having as possible values "macBook" or "sherlock". You can also create you own or modify the existing values Here you can find all the datasets that outlier scores have been pre-computed