• Assumption: a combination of multiple classifiers will improve classification performance. 假设：多个分类器的组合使用可以提升分类性能
    • General approach 常用方法
      • Given dataset Ensemble Methods - 图1, create Ensemble Methods - 图2 training datasets, Ensemble Methods - 图3. 给定数据集D，创建训练数据集 D1、D2…
      • Train a classification model Ensemble Methods - 图4 using training data set Ensemble Methods - 图5 (Ensemble Methods - 图6 使用训练集Di，训练分类模型Mi .
      • Combine Ensemble Methods - 图7 learned models (base classifiers), Ensemble Methods - 图8 to create a better model Ensemble Methods - 图9 结合所有的模型，成一个更好的模型。
    • Ensemble approaches can be distinguished by the training set creation methods训练集创建方法, training models训练模型创建方法, and combination methods组合模型. e.g.
      • Bagging
      • Boosting
      • Random Forest

    image.png
    An illustration of the ensemble approach. Ensemble Methods - 图11 is the original data. From the original data, an ensemble approach curates Ensemble Methods - 图12 training data sets, Ensemble Methods - 图13, and then trains Ensemble Methods - 图14 different classification models based on the training data sets. The combination of the trained models will be used for future prediction over new unlabelled data.
    image.png
    An example of the ensemble approach.
    (a) two lines represent decision boundaries of two different classification models. 两个不同分类器
    (b) An ensemble approach combines two decision boundaries to generate a more complex decision boundary.结合了两个不同的分类器