Machine learning | simple but feature standardization methods that can improve the effect of the model (comparison and analysis of robustscaler, minmaxscaler, standardscaler)

---

One 、 Method resolution

We usually use sklearn Of StandardScaler To standardize the features , But if your data contains outliers , Then the effect may not be good .

The method introduced in this article is called RobustScaler, It can obtain more robust feature scaling results . And StandardScaler Zoom different , Outliers are not included at all RobustScaler In calculation . So in data sets that contain outliers , It is more likely to scale to a near normal distribution .

StandardScaler The calculation formula is as follows （ It's also called standard deviation standardization Z-Score）：

x ′ = x − μ σ