ランダムフォレストを使って探索的データ分析

by Kan Nishida

Slide 1

Slide 1 text

ϥϯμϜϑΥϨετΛ࢖ͬͯ ୳ࡧతσʔλ෼ੳ

Slide 2

Slide 2 text

Presenter ੢ా צҰ࿠ co-founder/ceo Exploratory ུྺ Exploratory, Inc.ͰCEO݉νʔϑɾϓϩμΫτɾΦ ϑΟαʔΛ຿ΊΔ͔ͨΘΒɺ೔ຊͰͷσʔλαΠΤϯ εɾϒʔτΩϟϯϓɾτϨʔχϯά ɺάϩʔόϧͷσʔ λαΠΤϯεɾϒʔελʔɾτϨʔχϯάͳͲͷ৔Λ ௨ͯ͠σʔλαΠΤϯεͷීٴͱڭҭʹऔΓ૊Ήɻ @KanAugust

Slide 3

Slide 3 text

Exploratory Data Analysis with Random Forest

Slide 4

Slide 4 text

Exploratory Data Analysis with Random Forest

Slide 5

Slide 5 text

ܾఆ໦ σʔλ αϯϓϧ αϯϓϧ αϯϓϧ ౤ථ ౤ථ ౤ථ ݁࿦ … ϥϯμϜαϯϓϧ ౤ථׂ͕Εͨ৔߹ฏۉΛͱΓɺ ֬཰ͱͯ͠ѻ͏ ϥϯμϜϑΥϨετ

Slide 6

Slide 6 text

ϥϯμϜϑΥϨετͬͯɺ༧ଌͷͨΊ͡Όͳ͍ͷʁ

Slide 7

Slide 7 text

Employee Is_Salesrep Is_Male Attrition A TRUE FALSE TRUE B FALSE TRUE TRUE C TRUE FALSE FALSE D FALSE FALSE FALSE E TRUE TRUE TRUE Z TRUE TRUE ? ͜ͷैۀһ͸ࣙΊΔͰ͠ΐ͏͔ʁ ෼ྨ - ܾఆ໦

Slide 8

Slide 8 text

Is_Salesrep TRUE FALSE Is_Male TRUE FALSE … TRUE FALSE TRUE FALSE Z FALSE TRUE ? Employee Is_Salesrep Is_Male Attrition A TRUE FALSE TRUE B FALSE TRUE TRUE C TRUE FALSE TRUE D FALSE FALSE FALSE E FALSE TRUE TRUE ࣙΊΔʁ ࣙΊΔ ࣙΊͳ͍ ࣙΊΔ ෼ྨ - ܾఆ໦

Slide 9

Slide 9 text

… TRUE FALSE TRUE FALSE X TRUE TRUE ? Employee Is_Salesrep Is_Male Attrition A TRUE FALSE TRUE B FALSE TRUE TRUE C TRUE FALSE TRUE D FALSE FALSE FALSE E FALSE TRUE TRUE ࣙΊΔ ࣙΊͳ͍ ࣙΊΔ Is_Salesrep Is_Male ࣙΊΔʁ ෼ྨ - ܾఆ໦

Slide 10

Slide 10 text

TRUE FALSE TRUE FALSE … TRUE FALSE TRUE FALSE X TRUE TRUE TRUE Employee Is_Salesrep Is_Male Attrition A TRUE FALSE TRUE B FALSE TRUE TRUE C TRUE FALSE TRUE D FALSE FALSE FALSE E FALSE TRUE TRUE ࣙΊΔ ࣙΊͳ͍ ࣙΊΔ Is_Salesrep Is_Male ࣙΊΔʁ ෼ྨ - ܾఆ໦

Slide 11

Slide 11 text

Ͳͷม਺ΛઌʹධՁ͢Δ͔Ͱ༧ଌ݁Ռ͕ૣ͘ಘΒΕΔ

Slide 12

Slide 12 text

͜ͷ৘ใΛ΋ͱʹ ม਺ॏཁ౓(Variable Importance)͕ ಘΒΕΔ

Slide 13

Slide 13 text

RF Model

Slide 14

Slide 14 text

ม਺ॏཁ౓

Slide 15

Slide 15 text

Ͳͷม਺͕ଞͷม਺ʹൺ΂ͯॏཁ͔͸෼͔ͬͨ

Slide 16

Slide 16 text

͔͠͠… ͦͷม਺͕Ͳ͏Өڹ͍ͯ͠Δͷ͔͕Α͘෼͔Βͳ͍

Slide 17

Slide 17 text

Job Level

Slide 18

Slide 18 text

Monthly Income

Slide 19

Slide 19 text

ͦ͜Ͱ… edarf - EDA with RandomForest ύοέʔδ

Slide 20

Slide 20 text

No content

Slide 21

Slide 21 text

Partial Dependency

Slide 22

Slide 22 text

Job Level Monthly Income Stock Option 1 5000 1 3 10000 2 2 4000 1 2 6000 2 3 50000 3 … 1 5000 2 ݩͷσʔλ͔Β

Slide 23

Slide 23 text

Job Level Monthly Income Stock Option 1 5000 1 3 10000 2 2 4000 1 2 6000 2 3 50000 3 αϯϓϧ͢Δ

Slide 24

Slide 24 text

Job Level Monthly Income Stock Option 1 5000 1 1 10000 2 1 4000 1 1 6000 2 1 50000 3 Job Level͕̍ɼ̎ɼ̏ͷ̏ͭͩͱ͢Δͱɺ ͜ͷྻͷ஋͚ͩΛม͑ɺଞͷྻ͸ͦͷ··ͰɺγϛϡϨʔτ͢Δ Job Level Monthly Income Stock Option 2 5000 1 2 10000 2 2 4000 1 2 6000 2 2 50000 3 Job Level Monthly Income Stock Option 3 5000 1 3 10000 2 3 4000 1 3 6000 2 3 50000 3

Slide 25

Slide 25 text

Job Level Monthly Income Stock Option 1 5000 1 1 10000 2 1 4000 1 1 6000 2 1 50000 3 Job Level͕̍ͷ৔߹ RF Model Job Level Monthly Income Stock Option Prediction Probability 1 5000 1 0.3 1 10000 2 0.7 1 4000 1 0.2 1 6000 2 0.1 1 50000 3 0.6 0.38 ฏۉ஋

Slide 26

Slide 26 text

Job Level Monthly Income Stock Option 2 5000 1 2 10000 2 2 4000 1 2 6000 2 2 50000 3 Job Level Monthly Income Stock Option Prediction Probability Predict 2 5000 1 0.2 2 10000 2 0.3 2 4000 1 0.1 2 6000 2 0.1 2 50000 3 0.3 RF Model Job Level͕̎ͷ৔߹ 0.2 ฏۉ஋

Slide 27

Slide 27 text

Job Level Monthly Income Stock Option 3 5000 1 3 10000 2 3 4000 1 3 6000 2 3 50000 3 Job Level Monthly Income Stock Option Prediction Probability 3 5000 1 0.2 3 10000 2 0.3 3 4000 1 0.1 3 6000 2 0.1 3 50000 3 0.3 RF Model Job Level͕̏ͷ৔߹ 0.2 ฏۉ஋

Slide 28

Slide 28 text

ϓϩοτ͢Δ FALSE TRUE

Slide 29

Slide 29 text

with RStudio

Slide 30

Slide 30 text

with Exploratory

Slide 31

Slide 31 text

No content

Slide 32

Slide 32 text

σʔλαΠΤϯεɾϒʔτΩϟϯϓ τϨʔχϯά݄̏։࠵ https://exploratory.io/training-jp

Slide 33

Slide 33 text

࿈བྷઌ Email [email protected] Website https://ja.exploratory.io Twitter @ExploratoryData Facebook ExploratoryJapan