Data mining 2021-4-27 class notes

data mining

Supervised learning （ classification ）

Unsupervised learning （ clustering ）

prediction problems：Classification vs. Numeric Prediction

Test set data , It should be strictly separated from the training set . If you have seen your homework before the exam , You can't detect whether you really master

Decision Tree Induction（ Decision tree induction ）

intuitive , Easy to understand

Information gain , entropy

High entropy , High uncertainty
Low entropy , Low uncertainty

Continuous value ：<20, 20~30, 30 ~40, >40

Gain Ratio

Gini Index Used in CART

Overfitting And Tree Pruning
Somewhat branches In response anomalies perhaps noise perhaps outliers

The black line is better than the green line

Yes overfitting Two kinds of approaches

Prepruning:

threshold Difficult to choose

Postpruning:

Prune with another set of data （ Use more ）

Advantages of decision trees ：

1. Fast learning
2. It can be transformed into simple and understandable classification conditions
3. Easy to use SQL Realization
4. Acceptable classification accuracy

Bayes Classification Methods

An example of naive Bayesian classifier ：



The advantages of naive Bayes ：
Easy to implement , Excellent results
shortcoming ： Variables are often not independent

You can use Bayes Bayesian Belief Networks To partially solve the above problems