Background
It is a method of regression.
The key difference compared with LAD and OLS(Ordinary Least Squares) is that
- OLS 求”平均”:异常值通过平方放大,会强力”拉扯”回归线
- LAD 求”中位数”:异常值只贡献线性影响,对拟合线影响有限
Advantages/Disadvantages
优点:
- 对异常值鲁棒
- 适用于重尾分布数据
缺点:
- 解不唯一(可能有多个最优解)
- 计算复杂度高(需用线性规划求解,无闭式解)
- 对小误差不敏感(在原点不可微)