本文探讨了特征缩放（scaling）在数据预处理中的必要性，及其对不同机器学习算法的影响。特征缩放能消除变量尺度差异带来的偏差，在SVM和逻辑回归中效果显著，但对如随机森林、LGBM和XGBoost等决策树算法影响不大。原因在于决策树算法更关注类别变量间的差异，而非数值大小。
Scaling have different affection on regression or decisiontree
Scaling is a necessary step of preprocessing,it can help eliminating the bias caused by variable with different scales. It works in SVM or LR,but not in DecisionTree Algorithms like RF,LGBM and XGB.
The main reason that causes this difference is because DecisionTree algorithms’ working mechanism is to see how categorical variables differ,which is not like regression’s math issue. It does not rely on the scale of variables.
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
