Hyper Spectral Anomaly Detection (HSA): Advanced Anomaly Detection Across Disciplines

The Hyper Spectral Anomaly Detection (HSA) model is a statistics-based tool designed for unsupervised anomaly detection across various data types and applications. By leveraging graph theory, HSA analyzes relationships within a dataset on multiple time and length scales, making it effective for diverse fields such as cybersecurity, non-destructive materials testing, GIS tasks, and more, including both internal and external applications within the DOE system.

The HSA model operates by assessing the density and similarity of data points within a dataset, encoding this information into an affinity matrix. This matrix evolves to capture the data's structure on larger topographical scales. An anomaly score vector is generated and passed to a user-defined penalized objective function, which minimizes the anomaly scores. Data points with z-scores exceeding a specified threshold are flagged as anomalies. Additionally, HSA incorporates a multi-filter feature to reduce false positive rates, recording the indexes of predictions and creating a new dataset for further analysis.

This software is valuable for various sectors, including cybersecurity, where it helps narrow down internet protocol data to identify malicious events; non-destructive testing, where it detects material defects; GIS, for monitoring natural disasters and population development; nuclear applications, to identify undeclared behavior or material diversion; astronomy, to automate the study of star locations; and predictive maintenance, to flag potential equipment breakdowns. The HSA model enhances anomaly detection capabilities across disciplines, streamlining analysis and improving accuracy in identifying irregular data patterns.

This software is open-source and available at no cost. Access the software by visiting the GitHub link: https://github.com/IdahoLabUnsupported/Hyper_Spectral_Anomaly_Detection