SDO: outlier detection based on low density models

Last updated:

Dec 2017


If your are using any of the material below please cite the corresponding publication:


SDO (Sparse Data Observers) is an algorithm that scores data samples with estimations of distance-based outlierness. Alike other outlier detection algorithms, SDO is an eager learner that creates a low-density model of the dataset during a training phase and later compares new samples with the created model. Such scheme allows lightening the computational load during application phases, not requiring to recall old data samples again.

SDO is devised to be embedded in systems or frameworks that operate autonomously and must process large amounts of data in a continuos manner. SDO is a machine learning solution for Big Data and stream data applications.

Software, scripts, tools:

The MATLAB version can be downloaded from the SDO GitHub repository.

Datasets, experiments:

Example figures:

Two examples of SDO scoring for a 2-dimensional dataset (left) and a 3-dimensional dataset (rigth).