A Gradient Boosting Decision Tree Method based Air Combat Style Predication

Abstract

The combat style of the air target plays a key role in the air combat strategy and
combat result. If the combat style tendency of the air targets can be correctly predicted
during the combat, a counter strategy in a targeted manner can be adopted to greatly
improve the probability of winning the air combat. In the one-on-one air combat, the
basic combat style includes three kinds of style characteristics: the moderate, the
conservative and the radical. A new intelligent decision tree model based on the
Gradient Boosting Decision Tree (GBDT) method is proposed in the article, which
can effectively predict the combat style of target objects by using air combat
simulation confrontation behavior data. Firstly, the Classification and Regression Tree
(CART) is constructed to conduct Classification Regression and feature selection.
Then three different styles Artificial Intelligence?AI?of air combat behavior data
are generated by using evolutionary neural network technology. All parameters of
neural network strategy are then coded by Genetic Algorithm in real number. And by
designing different task target fitness functions, genetic evolution is carried out
according to the fitness value. The further neural network strategy with the best fitness
value is formed as the Artificial Intelligence?AI?for data collection. So the baseline
objects of the moderate, the conservative and the radical style are generated and a
large amount of confrontation date is obtained by fighting against each other in the air
combat simulation environment. Finally, the LightGBM framework is used to process
the large-scale confrontation data with a total training sample size of 180000.