Predictive Factor Analysis of Air-to-Air Engagement Outcomes Using Air Combat Manoeuvring Instrumentation Data.

Abstract

This study presents a novel predictive factor analysis of air-to-air engagement outcomes using a decade of air combat manoeuvring data (2009-2019) from the Air Combat Manoeuvring Instrumentation (ACMI) system of the Republic of Korea Air Force (ROKAF). The objective was to construct and evaluate an air-to-air combat hit prediction model using the ACMI system data to identify the critical factors influencing engagement outcomes. This methodology encompasses data preprocessing, feature engineering, binary classification model development, and model interpretation. This study utilises 17 features, including the attitude and speed of both aircraft, along with five additional features derived from the domain knowledge of the relative positions of the two aircraft. Four machinelearning algorithms were employed: logistic regression, random forest, XGBoost, and CatBoost. The best-performing model achieved an accuracy of 83.0 %, noticeably outperforming the baseline at 76.2 %. The analysis revealed that positional information is more crucial than attitude information in predicting engagement outcomes, with the spatial separation between aircraft emerging as the most influential factor. This study showcasings a standard procedure for utilising ACMI system data and demonstrating the effectiveness of machine learning in analysing air combat data.
Subjects
FACTOR analysis; RANDOM forest algorithms; MACHINE learning; LOGISTIC regression analysis; AIR forces

Description