Michal Dichter, Applied Technology Associates, a BlueHalo Company
Keywords: Machine Learning, Launch Threat Assessment, Similarity-Based Classification
Abstract:
The transformation of real-time data into actionable information is a fundamental requirement for the fast and accurate threat assessment of both orbital and suborbital launches. The Similarity-Based Launch Classification Tool (SLCT) uses a novel approach based on machine learning to classify two target variables, rocket type and target regime, from a discrete stream of state-vector observations. At the heart of the SLCT is a subsequence search algorithm that uses dynamic time warping (DTW) to measure the distance between a variable-length query sequence and fixed-length reference sequence. Armed with DTW as a time series comparison method, Monte Carlo cross-validation is used to optimize the accuracy of a k-nearest-neighbors (k-NN) classifier over a range of observation time intervals. Several run-time optimizations and approximate methods are proposed to accelerate the subsequence search.
Date of Conference: September 14-17, 2021
Track: Machine Learning for SSA Applications