數(shù)據(jù)挖掘2015最新精品課程完整(第9講)-基于規(guī)則的分類方法-2課件

1、基于規(guī)則的分類基于規(guī)則的分類器（Rule-Based Classifier）使用一系列“ifthen” 對數(shù)據(jù)集分類規(guī)則: (Condition) y此處 Condition 為多個屬性-值對的合取 y 為類標(biāo)號LHS(IF 部分): 規(guī)則前件或前提RHS(then 部分): 規(guī)則結(jié)論分類規(guī)則例子: (血的類型=溫血) (下蛋=Yes) 鳥 (稅收收入 Bird規(guī)則 R3 覆蓋 grizzly bear = Mammal規(guī)則的沖突觸發(fā)：規(guī)則被滿足激活：該規(guī)則為唯一滿足的R1: (Give Birth = no) (Can Fly = yes) BirdsR2: (Give Birth = no

2、) (Live in Water = yes) FishesR3: (Give Birth = yes) (Blood Type = warm) MammalsR4: (Give Birth = no) (Can Fly = no) ReptilesR5: (Live in Water = sometimes) Amphibians A lemur triggers rule R3, so it is classified as a mammalA turtle triggers both R4 and R5A dogfish shark triggers none of the rules沖

3、突解決規(guī)模序(size ording): 要求最嚴(yán)格的規(guī)則賦予最高優(yōu)先級 (i.e., 最多屬性測試)基于類的序: 按照類的頻繁性或錯分代價的降序排列基于規(guī)則的序 (決策表): 根據(jù)規(guī)則的質(zhì)量度量或?qū)＜乙庖?，?guī)則組織為長的優(yōu)先級列表有序規(guī)則集根據(jù)優(yōu)先權(quán)對規(guī)則進行排序?qū)σ粋€待分類的記錄若滿足多條規(guī)則，則使用排在最前面的對其進行分類。若不滿足任何規(guī)則，則使用默認類別。R1: (Give Birth = no) (Can Fly = yes) BirdsR2: (Give Birth = no) (Live in Water = yes) FishesR3: (Give Birth = yes) (

4、Blood Type = warm) MammalsR4: (Give Birth = no) (Can Fly = no) ReptilesR5: (Live in Water = sometimes) Amphibians 構(gòu)造分類規(guī)則直接方法: 直接從數(shù)據(jù)中提取規(guī)則 e.g. FOIL, AQ, CN2, RIPPER 間接方法: 從其它分類模型中提取規(guī)則 、e.g. decision trees, neural networks, etc示例順序覆蓋的要點產(chǎn)生規(guī)則消除實例規(guī)則評價停止標(biāo)準(zhǔn)規(guī)則的剪枝消除實例不消除實例?不消除正例?不消除負例?總是產(chǎn)生同一條規(guī)則可能過高估計后續(xù)規(guī)則的正確度

5、可能過低估計后續(xù)規(guī)則的正確度規(guī)則評價標(biāo)準(zhǔn):AccuracyLaplaceM-estimaten : Number of instances covered by rulenc : Number of instances corrected classified by rulek : Number of classesp : Prior probability停止標(biāo)準(zhǔn)停止標(biāo)準(zhǔn)計算增益若增益不顯著，則舍棄新規(guī)則規(guī)則剪枝與決策樹的后剪枝相似降低錯誤剪枝: 刪去規(guī)則的一個合取支(conjunct)在測試集上比較剪枝前后的錯誤率若出錯率降低，則剪掉這一合取支直接方法的總結(jié)產(chǎn)生一條單一規(guī)則刪除規(guī)則覆蓋到的元組若需要，對規(guī)則進行剪枝將規(guī)則添加到當(dāng)前的規(guī)則集中重復(fù)，直到滿