Skip to main content

Rule-Based Classification 基於規則的分類

5.5 Decision Rules | Interpretable Machine Learning

It is good: Rule Based Classification | An Explorer of Things

从零开始-Machine Learning学习笔记(31)-规则学习_first-order inductive learner-CSDN博客

  • Conjunct
  • Antecedent
  • Consequence

Building Rules​

Direct Method

  • CN2
  • RIPPER

Indirect Method

  • C4.5 (extract rules from an unpruned decision tree)

Class-based ordering 基於類別的排序​

Decreasing order of prevalence or misclassification cost per classs

Sequential covering 順序覆蓋​

  1. Add Rule
  2. Remove cover data point
  3. Loop (1) (2)

例如 CN2 和 RIPPER

CN2 use Specific-to-General Strategy 特定到通用策略

RIPPER use General-to-Specific Strategy 通用到特定策略

PART (partial tree) decision list​

Generate rule from decision tree

ZeroR​

ZeroR-baseline分类器 - 愚人_同乐 - 博客园

Just Y = majority class

OneR​

5.5.1 Learn Rules from a Single Feature (OneR)

First-Order Inductive Learner (FOIL) Information Gain​

:: KDnet - Find information and resources on Machine Learning, Knowledge Discovery, Data Mining, Case-based Reasoning, and Knowledge Acquisition ::

FOIL_Gain=p′ (logp′p′+n′−logpp+n)\text{FOIL\_Gain} = p' \text{ } (log \frac{p'}{p' + n'} - log \frac{p}{p + n})

FOIL_Prune​

FOIL_Prune=p−np+n\text{FOIL\_Prune} = \frac{p - n}{p + n}

RIPPER​

RIPPER (Repeated Incremental Pruning to Produce Error Reduction)

Ripper Algorithm - GeeksforGeeks

RIPPER (重複增量修剪以減少錯誤)是順序覆蓋演算法的變體

RIPPER 更為複雜，它使用後處理階段（規則修剪）來最佳化決策清單（或集合）

RIPPER 可以在有序或無序模式下運行，並產生決策清單或決策集