The document discusses bagging ensemble techniques. It explains that bagging aims to reduce variance by training multiple classifiers on randomly sampled subsets of the training data and combining their predictions through majority voting. This technique is effective for models that have high variance and low bias by generating diverse models that stabilize the prediction.

1. Bagging ensemble
1

2. Background knowledge
 Bias(偏差)：describe the difference(or distance) between
predicted result and real result。
 Variance(變異)：describe the instability of predicted result。
2
準是 bias 小，確是 variance 小，準確是相對概念，因為
bias-variance tradeoff。

3. Background knowledge
 Bias-Variance Tradeoff(偏差和變異之權衡)
 Underfitting：when model’s predicted result with high bias, the error of training
set and validation set are very high.
 Overfitting： when model’s predicted result with high variance
the error of training set is very low but validation set is very high.
3

4. How to deal with?
 How to deal with underfitting：
1. Increasing model iteration times.
2. Using more powerful model to replace.
3. Generate more feature to use.
4. Decreasing regularization.
 How to deal with underfitting：
1. Increasing training set.
2. Decreasing training feature.
3. Increasing regularization.
4

5. Conception
 Bagging ensemble is first using bootstrap to select training subset randomly
secondly using those training subset to build classifiers, finally using majority
vote to get final result.

6. Conception
1. Mainly to reduce Variance，have a little effect on Bias
2. Models for High Variance & Low Bias