Upgrade to Pro — share decks privately, control downloads, hide ads and more …

了解决策树和C4.5算法

Avatar for yafei002 yafei002
January 08, 2017
Technology
1
260

 了解决策树和C4.5算法

Avatar for yafei002

yafei002

January 08, 2017
Tweet

More Decks by yafei002

See All by yafei002
了解人工神经网络
Avatar for yafei002 yafei002
1
210
了解朴素贝叶斯
Avatar for yafei002 yafei002
1
230
了解K-Means算法
Avatar for yafei002 yafei002
1
210
了解KNN算法
Avatar for yafei002 yafei002
0
170
数据可视化之视觉感知与认知
Avatar for yafei002 yafei002
1
380
数据可视化之地理信息可视化
Avatar for yafei002 yafei002
1
390
数据可视化之层次和网络数据可视化（上）
Avatar for yafei002 yafei002
1
660
数据可视化之复杂高维多元数据的可视化（上）
Avatar for yafei002 yafei002
1
290
Data Visualization Introduction and History
Avatar for yafei002 yafei002
1
320

Other Decks in Technology

See All in Technology
_第4回__AIxIoTビジネス共創ラボ紹介資料_20251203.pdf
Avatar for AIxIoTビジネス共創ラボ iotcomjpadmin
0
170
Oracle Database@Azure：サービス概要のご紹介
Avatar for oracle4engineer oracle4engineer
PRO
3
260
Introduction to Sansan Meishi Maker Development Engineer
Avatar for Sansan, Inc. sansan33
PRO
0
330
Agentic AIが変革するAWSの開発・運用・セキュリティ ～Frontier Agentsを試してみた～ / Agentic AI transforms AWS development, operations, and security I tried Frontier Agents
Avatar for Yuji Oshima yuj1osm
0
190
「エッジ×分散生成AI」の技術と変わる産業、そしてITの未来
Avatar for piacerex piacerex
0
110
Eight Engineering Unit 紹介資料
Avatar for Sansan, Inc. sansan33
PRO
0
6.1k
Introduction to Sansan for Engineers / エンジニア向け会社紹介
Avatar for Sansan, Inc. sansan33
PRO
5
59k
Data Hubグループ 紹介資料
Avatar for Sansan, Inc. sansan33
PRO
0
2.5k
Digitization部 紹介資料
Avatar for Sansan, Inc. sansan33
PRO
1
6.4k
20251203_AIxIoTビジネス共創ラボ_第4回勉強会_BP山崎.pdf
Avatar for AIxIoTビジネス共創ラボ iotcomjpadmin
0
170
AIと融ける人間の冒険
Avatar for pujisi pujisi
0
110
Everything As Code
Avatar for わいわい yosuke_ai
0
480

Featured

See All Featured
Bootstrapping a Software Product
Avatar for Garrett Dimon garrettdimon
PRO
307
120k
Visual Storytelling: How to be a Superhuman Communicator
Avatar for David Neal reverentgeek
2
400
Being A Developer After 40
Avatar for Adrian Kosmaczewski akosma
91
590k
Organizational Design Perspectives: An Ontology of Organizational Design Elements
Avatar for Dr. Kim W Petersen kimpetersen
PRO
0
48
The Organizational Zoo: Understanding Human Behavior Agility Through Metaphoric Constructive Conversations (based on the works of Arthur Shelley, Ph.D)
Avatar for Dr. Kim W Petersen kimpetersen
PRO
0
210
Reality Check: Gamification 10 Years Later
Avatar for Sebastian Deterding codingconduct
0
2k
Git: the NoSQL Database
Avatar for Brandon Keepers bkeepers
PRO
432
66k
Helping Users Find Their Own Way: Creating Modern Search Experiences
Avatar for Dan Newman danielanewman
31
3.1k
Building a Scalable Design System with Sketch
Avatar for Laura Van Doore lauravandoore
463
34k
Building AI with AI
Avatar for Ines Montani inesmontani
PRO
1
600
Reflections from 52 weeks, 52 projects
Avatar for Jefferson Lam jeffersonlam
355
21k
Put a Button on it: Removing Barriers to Going Fast.
Avatar for kastner kastner
60
4.1k

Transcript

  1. 了解决策树与 C4.5算法 yafei002

  2. 决策树工作原理

  3. 如何选择节点 熵（entropy）：信息的期望值

  4. 如何选择节点 决策树算法选择最大增益作为最佳划分 增益：

  5. 如何选择节点 增益偏向于取值多的属性 增益率（Gain ration）将输出节点的个数纳入考量 C4.5

  6. 决策树剪枝 为了避免生成的树过多从而过度拟合训练数据，需要对生成的决策树进行剪枝。 C4.5算法引入了悲观剪枝的方法。 悲观剪枝： 1. 一个节点对应N个实例和E个错误，则该 节点的经验错误率=（E+罚项）/ N 2. 一个子树有L个叶子节点，这些叶子节点共包含∑N个实例和∑E个错误，则该

    子树的经验错误率=（ ∑ E+L*罚项）/ ∑N 3. 假设子树被它的最佳叶节点替换后，在训练数据集的错误为J，如果 J+罚项<= ∑E+罚项*L + （ ∑ E+罚项）的一倍的标准差 则决定用该最佳节点替换子树

  7. 决策树剪枝 X T1 T2 T3 X T1 T2 T3 T2

    （X输出最大） （最佳叶节点） 剪枝是单一的自底向上的遍历的过程 图：一个剪枝的中间步骤

  8. PANG-NINGTAN, MICHAELSTEINBACH, & VIPINKUMAR. (2011). 数据挖掘导 论:完整版. 人民邮电出版社. 吴信东, &

    VipinKumar. (2013). 数据挖掘十大算法. 清华大学出版社. 参考资料