Chapter 12 Knowing when to worry Hazard Functions

  • Slides: 76
Download presentation
Chapter 12 Knowing when to worry: Hazard Functions and Survival Analysis in Marketing 危險函數及存活分析

Chapter 12 Knowing when to worry: Hazard Functions and Survival Analysis in Marketing 危險函數及存活分析 林家如 張惠盈 國立中興大學行銷研究所 8 June 2006 Hazard Functioin and Survival Analysis 1

報告流程 a. b. c. d. e. f. g. h. 專有名詞介紹 存活分析之基本概念 顧客維持( Customer Retention)

報告流程 a. b. c. d. e. f. g. h. 專有名詞介紹 存活分析之基本概念 顧客維持( Customer Retention) 危險 (Hazards) 由“危險”推導“存活” (From Hazard to Survival) 比例危險 (Proportional Hazards) 課程回顧 SAS操作示範 8 June 2006 Hazard Functioin and Survival Analysis 2

專有名詞介紹 a. n n Customer Tenure 顧客資歷 Customer Lifetime 顧客生命 n Median customer lifetime

專有名詞介紹 a. n n Customer Tenure 顧客資歷 Customer Lifetime 顧客生命 n Median customer lifetime n Customer half-life 顧客生命之中位數 n Average customer lifetime 平均顧客生命 Customer Retention 顧客維持 Retention Curve 維持曲線 Hazard 危險 Censoring 設限 8 June 2006 Hazard Functioin and Survival Analysis 3

4. 平均顧客資歷 Average Tenure n Average tenure =Average remaining lifetime =Truncated Mean lifetime n

4. 平均顧客資歷 Average Tenure n Average tenure =Average remaining lifetime =Truncated Mean lifetime n 無法計算「真正」的平均生命,除非所有的顧 客均離開。因此特定時間內的平均生命,統計 學家稱為截距式平均生命 8 June 2006 Hazard Functioin and Survival Analysis 9

n 平均顧客資歷: 特定期間內,位於維持曲線下方的面積 10 1/10 8 June 2006 20 20/10 Hazard Functioin and Survival

n 平均顧客資歷: 特定期間內,位於維持曲線下方的面積 10 1/10 8 June 2006 20 20/10 Hazard Functioin and Survival Analysis 10

危險 Hazards d. 屬無母數(non-parametric)之統計方法 可延伸出更有用之存活曲線 1. 2. 3. 危險機率 Hazard Probability 危險函數 Hazard Function

危險 Hazards d. 屬無母數(non-parametric)之統計方法 可延伸出更有用之存活曲線 1. 2. 3. 危險機率 Hazard Probability 危險函數 Hazard Function 設限 Censoring 8 June 2006 Hazard Functioin and Survival Analysis 13

Example: 圖示+計算 1 2 3 4 5 n白圈圈: 顧客已經離開,確實tenure已知 n黑圈圈: 顧客尚存活,確實tenure未知 , 必須設限 8

Example: 圖示+計算 1 2 3 4 5 n白圈圈: 顧客已經離開,確實tenure已知 n黑圈圈: 顧客尚存活,確實tenure未知 , 必須設限 8 June 2006 Hazard Functioin and Survival Analysis 6 7 19

Custo Censore Teun mer d re Time 0 Time 1 Time 2 Time 3

Custo Censore Teun mer d re Time 0 Time 1 Time 2 Time 3 Time 4 Time 5 1 Y 5 Active Active 2 N 4 Active Stopped Censored 3 N 3 Active Stopped Censored 4 Y 3 Active Censored 5 N 2 Active Stopped Censored 6 Y 1 Active Censored 7 N 1 Active Stopped Censored Time 0 Time 1 Time 2 Time 3 Time 4 Time 5 Active 7 6 4 3 1 1 Stopped 0 1 1 0 Censored 0 0 2 3 5 6 Hazard 8 June 2006 0/(0+7 1/(1+6 1/(1+4 1/(1+3 1/(1+1 0/(0+1 )=0% )=14% )=20% )=25% )=50% )=0% Hazard Functioin and Survival Analysis 20

被強制驅離的顧客,應標示Censored而不是Stopped 8 June 2006 Hazard Functioin and Survival Analysis 22

被強制驅離的顧客,應標示Censored而不是Stopped 8 June 2006 Hazard Functioin and Survival Analysis 22

From Hazard to Survival 由“危險”推導“存活” 1. 2. 維持曲線 由危險函數推導更為精準的存活曲線 8 June 2006 Hazard Functioin

From Hazard to Survival 由“危險”推導“存活” 1. 2. 維持曲線 由危險函數推導更為精準的存活曲線 8 June 2006 Hazard Functioin and Survival Analysis 23

存活曲線(Survival Curve) 較平滑 較精準 逐漸下降 8 June 2006 Hazard Functioin and Survival Analysis 27

存活曲線(Survival Curve) 較平滑 較精準 逐漸下降 8 June 2006 Hazard Functioin and Survival Analysis 27

比例危險 Proportional Hazards p p 範例 分層(Stratification) Cox比例危險回歸 比例危險的限制 8 June 2006 Hazard Functioin

比例危險 Proportional Hazards p p 範例 分層(Stratification) Cox比例危險回歸 比例危險的限制 8 June 2006 Hazard Functioin and Survival Analysis 28

存活分析之應用 p p 處理不同型態的離開(Attrition) 顧客何時回來 預測 隨時間改變的危險 8 June 2006 Hazard Functioin and Survival

存活分析之應用 p p 處理不同型態的離開(Attrition) 顧客何時回來 預測 隨時間改變的危險 8 June 2006 Hazard Functioin and Survival Analysis 42

n Tenure=1,並在 2003年停止的顧客風險人口 n 在 2002年 12月31日與2002年 12月30日之間開始的顧 客,可能有一天的資歷,並在 2003年停止 Tenure=1 12/30 8 June

n Tenure=1,並在 2003年停止的顧客風險人口 n 在 2002年 12月31日與2002年 12月30日之間開始的顧 客,可能有一天的資歷,並在 2003年停止 Tenure=1 12/30 8 June 2006 12/31 Hazard Functioin and Survival Analysis 57

n Tenure=365,並在 2003年停止的顧客風險人口 2002/1/1 8 June 2006 2002/12/31 Hazard Functioin and Survival Analysis Tenure=365

n Tenure=365,並在 2003年停止的顧客風險人口 2002/1/1 8 June 2006 2002/12/31 Hazard Functioin and Survival Analysis Tenure=365 58

SAS操作示範 p p 客戶資料說明 操作示範 8 June 2006 Hazard Functioin and Survival Analysis 63

SAS操作示範 p p 客戶資料說明 操作示範 8 June 2006 Hazard Functioin and Survival Analysis 63

客戶資料說明 n Variables n n n n CUSTOMER_ID:行動電話號碼 RATE_PLAN:費率(高、中、低) MONTHLY_FEE:每月電話費 MARKET:地點 CHANNEL:通路 START_DATE:起租日 CANCEL_DATE:終止日

客戶資料說明 n Variables n n n n CUSTOMER_ID:行動電話號碼 RATE_PLAN:費率(高、中、低) MONTHLY_FEE:每月電話費 MARKET:地點 CHANNEL:通路 START_DATE:起租日 CANCEL_DATE:終止日 CANCEL_TYPE:N活躍、V自願離去、I被迫離去 8 June 2006 Hazard Functioin and Survival Analysis 64

操作示範 n 新建立Library時,勾選Enable at startup 8 June 2006 Hazard Functioin and Survival Analysis 65

操作示範 n 新建立Library時,勾選Enable at startup 8 June 2006 Hazard Functioin and Survival Analysis 65

n Input Data Source 8 June 2006 Hazard Functioin and Survival Analysis 66

n Input Data Source 8 June 2006 Hazard Functioin and Survival Analysis 66

n Open Input Data Source Node 選擇 RAW 8 June 2006 選擇整 筆資料 Hazard

n Open Input Data Source Node 選擇 RAW 8 June 2006 選擇整 筆資料 Hazard Functioin and Survival Analysis 67

n 增加SAS Code Node 8 June 2006 Hazard Functioin and Survival Analysis 68

n 增加SAS Code Node 8 June 2006 Hazard Functioin and Survival Analysis 68

n n Program>Score 輸入設限指令 n n 尚存活的顧客tenure=20061231 -起租日,標注設限 其他:終止日-起始日,不設限 8 June 2006 Hazard Functioin

n n Program>Score 輸入設限指令 n n 尚存活的顧客tenure=20061231 -起租日,標注設限 其他:終止日-起始日,不設限 8 June 2006 Hazard Functioin and Survival Analysis 69

n Export增加score選項 8 June 2006 Hazard Functioin and Survival Analysis 70

n Export增加score選項 8 June 2006 Hazard Functioin and Survival Analysis 70

n 增加Score Node n n 執行SAS Code進行設限 增加tenure, censored二個變數 8 June 2006 Hazard Functioin

n 增加Score Node n n 執行SAS Code進行設限 增加tenure, censored二個變數 8 June 2006 Hazard Functioin and Survival Analysis 71

n Data>對應到資料 8 June 2006 Hazard Functioin and Survival Analysis 72

n Data>對應到資料 8 June 2006 Hazard Functioin and Survival Analysis 72

n Setting>Apply training data score code to score data set 8 June 2006 Hazard

n Setting>Apply training data score code to score data set 8 June 2006 Hazard Functioin and Survival Analysis 73

n 執行score node,新增的兩個變數已經加入 8 June 2006 Hazard Functioin and Survival Analysis 74

n 執行score node,新增的兩個變數已經加入 8 June 2006 Hazard Functioin and Survival Analysis 74

n n n 再新增一個SAS CODE 對應到Score的資料 Program>run(執行SAS Process) n n n proc lifetest data=emdata.

n n n 再新增一個SAS CODE 對應到Score的資料 Program>run(執行SAS Process) n n n proc lifetest data=emdata. sd_k 8 xbw outsurv=lifetest method=act intervals= 0 to 1000 by 1 plots=(s, h); time tenure*censored(1); run; 8 June 2006 Hazard Functioin and Survival Analysis 75

n Survival Curve 8 June 2006 Hazard Functioin and Survival Analysis 76

n Survival Curve 8 June 2006 Hazard Functioin and Survival Analysis 76

n Hazards Curve 8 June 2006 Hazard Functioin and Survival Analysis 77

n Hazards Curve 8 June 2006 Hazard Functioin and Survival Analysis 77

The end Thank you for your listening 8 June 2006 Hazard Functioin and Survival

The end Thank you for your listening 8 June 2006 Hazard Functioin and Survival Analysis 78