基于随机森林的目标意图识别

Target intention recognition based on random forest

运用机器学习方法及时准确识别目标作战意图,对于智能化战争中的军事指挥决策十分重要。相较于其他目标意图识别方法,随机森林算法具有抗噪声性能良好、数据集适应力强、训练速度快和实现简单等优点。采用集成学习思路,基于随机森林建立目标意图识别模型,并选用2015年全国研究生数学建模竞赛A题中已知意图的15批空中目标数据,通过WEKA软件提供的随机森林算法分析构造随机决策树,使用留一法检验识别性能,选取合适的算法参数,分析判断出未知意图的12批空中目标的作战意图。运算结果表明随机森林在测试集上的识别精度为83%,高于通过留一法验证的精度,也高于其他6支参赛一等奖获奖队伍的精度。最后通过与其他参赛获奖队伍算法的结果进行对比,逐一分析不同算法对不同意图的精度和召回率,寻找误差原因,得出随机森林是一种简便、快速、高效算法的结论,其识别准确度较其他算法具有一定的优势。

It is important for military command decisions in intelligent warfare to identify target combat intention timely and accurately by machine learning method. In comparison with other target intention recognition methods,the random forest(RF)algorithm has the advantages of good anti-noise performance,strong adaptability of data sets,fast training speed and simple implementation. The integrated learning idea is adopted to establish a target intention recognition model based on RF. The data of 15 batches of air targets with known intentions, which is selected from question A of the 2015 National Graduate Mathematical Modelling Competition,are adopted. A random decision tree is constructed by analyzing the RF algorithm provided by WEKA(Waikato environment for knowledge analysis). The leave-one-out method is used to check the recognition performance and the appropriate algorithm parameters are adopted to analyze and judge the combat intention of the 12 batches of air targets with unknown intention. The operation results show that the recognition accuracy of RF on the test set is 83%,which is higher than that verified by the leave-one-out method,and also higher than that of the algorithms used by the other six teams winning the first prize.The results are contrasted with those got by the algorithms used for the other winning teams. The accuracy and recall rate of the different algorithms for different intentions were analyzed one by one to find out the causes of the errors. It is concluded that the RF is a simple,fast and efficient algorithm,and its recognition accuracy has certain advantages over other algorithms.