W. A. Yousef, I. Traore, and W. Briguglio: "Unsupervised and nonparametric approach for visualizing outliers and invariant detection scoring", US Patent 63168686 (pending).
W. A. Yousef, I. Traoré and W. Briguglio, "UN-AVOIDS: Unsupervised and Nonparametric Approach for Visualizing Outliers and Invariant Detection Scoring," in IEEE Transactions on Information Forensics and Security, vol. 16, pp. 5195-5210, 2021, doi: 10.1109/TIFS.2021.3125608.github.com/isotlaboratory/UNAVOIDS-Code codeocean.com/capsule/2122579/tree/v1
UNAVOIDS: Unsupervised and Nonparametric Approach for Visualizing Outliers and Invariant Detection Scoring
VAAD is a data visualization software that aids the analyst in detecting outliers through exploration by providing UNAVOIDS, the patented visualization plot for anomaly detection, along with several other plots, such as matrix plots, parallel coordinates, etc., in a single interactive visualization environment, where linking and brushing are provided.
Through the limited trial online version on this site.
Through the PyPi package that you can integrate with your system (needs a quotation for commerical use)
Through re-implementing UN-AVOIDS, where you can start from the plain code (needs a quotation for commercial use)
It is quite important to emphasize that the utility of our VAAD, along with its internal algorithm of anomaly detection, is not explicit only to the network security domain from which the word anomaly is borrowed. It is applicable to any dataset that is required to detect those observations that look strange (anomalous) to the rest of the population. Therefore, it is applicable to any domain from which datasets are acquired. Few examples, out of other possible dozens, are listed below:
Network Security is a natural domain to apply anomaly detection. Any activity that is considered as a network attack will have a feature vector that looks strange (anomalous) if compared to other network activities.
Disease Detection in wide range of medical applications fits into the same category of anomalous detection. For example, any type of breast cancer (mass, microcalcification, bilateral asymmetry, etc.) appears as tissues of some sort of abnormality (anomaly) in mammograms.
Fraud Detection of financial institutions transactions is another example of anomaly detection. All transactions that do not look normal to the rest of transaction population should raise a flag least it is a fraud.
The full details appear in the IEEE publication . In summary, the VAAD's component that detects anomalies is UN-AVOIDS, which enjoys the following features:
unsupervised: where detection is pursued directly with no prior training. It does not assume any data labeling for any subset of the data, which implies that even single-class labeling is not assumed, as opposed to some other algorithms.
nonparametric: in two senses: (1) It does not assume, or take as an input, a probability distribution of the data. (2) It is parameter free, where there is no tuning or adjusting parameter that controls the accuracy of the algorithm.
an approach for visualizing outliers: where the analyst can explore the dataset and visually detect the outliers. Moreover, the visualization can be provided through any data visualization software as an interactive plot that is linked to other interactive plots. Such software enables the analyst to perform elaborate data visualization exploration and manipulations.
an invariant detection scoring: where the detection algorithm provides a score that is normalized to [0,1], which expresses the level of anomalousness (or outlierness) of an observation, as opposed to a hard-threshold decision.