web数据挖掘__3监督学习1.pptxVIP

Chapter 3: Supervised Learning;Road Map;An example application;Another application;Machine learning and our focus;The data and the goal;An example: data (loan application);An example: the learning task;Supervised vs. unsupervised Learning;Supervised learning process: two steps;What do we mean by learning?;An example;Fundamental assumption of learning;Road Map;Introduction;The loan data (reproduced);A decision tree from the loan data;Use the decision tree;Is the decision tree unique?;From a decision tree to a set of rules;Algorithm for decision tree learning;Decision tree learning algorithm;Choose an attribute to partition data ;The loan data (reproduced);Two possible roots, which is better?;信息论原理;信道模型;不确定性;不确定性;互信息;信息熵定义;举例;讨论;条件熵定义;平均互信息量定义;信息论在决策树中的应用-属性选择;信息增益（ Information Gain）;信息论在决策树中的应用;决策树的学习过程;决策树的学习过程;举例;属性 Outlook Temperature Humidity Windy 类;属性 Outlook Temperature Humidity Windy 类;在决策树方法中，所要做的工作就是构造决策树将数据进行分类。因初始时属于P类和N类的实例个数均为12，故初始熵值为：;(2) 若选择Temp作为测试属性，其条件熵为：;(4) 若选择Windy作为测试属性，其条件熵为：;Humidity;结论;训练集(练习);;使用信息增益进行属性选择;Decision Tree (结果输出);Handling continuous attributes;An example in a continuous space;避免过学习、过度拟合;过度拟合（Overfiting);导致过度拟合的原因;An example;Avoid overfitting in classification;Road Map;Evaluating classification methods;Evaluation methods;Evaluation methods (cont…);Classification measures;Table of Confusion 混合矩阵;Example;TP,TN,FP,FN;分类的评价准则;分类的评价准则;An example;分类的评价准则;Road Map;Three approaches;Using Class Association Rules;Decision tree vs. CARs;There are many more rules;Decision tree vs. CARs (cont …);Building classifiers;CBA: Rules are sorted first;Classifier building using CARs;Using rules as features;Using normal association rules for classification