基于可调权重距离的最小一乘回归方法

Least Absolute Deviation Regression Base on Adjustable Weighting Distance

不同时期数据对因变量的影响权重不同,若不考虑数据的时间特性而构建回归模型,可能不一定得到最佳的回归模型.针对经济领域中数据时间跨度大,样本少以及可能存在异常点的特点,提出基于可调权重距离的最小一乘回归方法.建立了可调权重距离的权重系数确定方法,并给出基于MATLAB的模型求解方法.通过某船舶使用费用预测的应用,表明通过方法构建的模型具有更高的精度,值得借鉴.

Data from the different periods have dissimilar influence to dependent variable.The best regression model may not be gained if not taking into account the influencing weight-ing of the different periods data. For large data time span, the small sample size and the possi-ble existence of outliers, the least absolute deviations regression based on adjustable weightingdistance is put forward. The method of determining the weighting coefficient based on ad-justable weighting distance is made, and the solution method of the model based on Matlabis given. The result of the application instance on predicting the operation costs in a shipshows that the model founded by this method has better higher precision, which is worthy of referrence.