Intelligence Extraction Using Machine Learning for Threat Identification Purposes : An Overview

Abstract: Radar is an invaluable tool for detecting and assessing threats on land, on the seas and in the air. To properly evaluate threats, radar operators construct threat libraries where the signal characteristics of emitters are stored and mapped to specific types of platforms. In this project, methods for constructing these threat detection libraries from data obtained during real-life scenarios are investigated. A number of machine learning approaches are investigated and validated using general and method specific scoring methods. Using density based clustering methods and non-linear data transformation it is shown that Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) and spatial consistency metrics can be used to deinterleave and group signals to radar trace emitting platforms, from which suitable library parameters can be extracted. The results show that traditional metrics for evaluating cluster methods are not suited for evaluating data containing spatial information.