摘要
针对采用参数法进行武器系统费用估算时,往往需要从大量的特征中选择最优的特征子集,而人工选择特征的方法费时费力的问题,采用粒子群优化算法对支持向量机进行封装的方法选择特征子集,在建模过程中,将特征变量以及支持向量机模型的参数共同组合成待优化的自变量集,采用一体化的优化方法,同时确定最优的特征子集和支持向量机模型的参数;在选择优化目标时,为有效避免出现"过拟合"的情况,将模型的5折交叉验证的预测精确度作为适应度函数,并对实际问题进行验证。结果表明,该方法可以有效滤除无关特征,提高预测精确度。
In the process of parametric cost estimation,the optimum feature subset need to be chosen from substantive features.Artificial selection methods always need great time and effort.Particle swarm optimization（PSO） was adopted to implement the wrapper feature subset selection and then support vector machine method was used to establish the cost estimation model.In the process of model establishment,the independent variable set was composed of support vector machine model parameters and characteristic variables,and the optimum feature subset and support vector machine model parameters were obtained by adopting an integrative optimistic method with PSO.At the same time,in order to avoid over fitting,the five-fold cross validation forecast accuracy of the model was ascertained as the fitness function and verified aim to the practical problem.The calculation results showed the method could filter some irrelevant features and improve the forecast accuracy.
出处
《武汉理工大学学报（信息与管理工程版）》
CAS
2010年第6期994-997,共4页
Journal of Wuhan University of Technology：Information & Management Engineering
关键词
特征选择
支持向量机
粒子群优化算法
交叉验证
费用估算
feature selection
support vector machine
particle swarm optimization
cross validation
cost estimation
