data mining and predictive modeling for...
TRANSCRIPT
1
Data Mining and Predictive Modeling for HRData Mining Techniques for Analyzing Voluntary Turnover
Jason [email protected]
2
Agenda
IntroductionBackground and experience.
What is Data Mining?High level overview.
Decision Tree ExamplesOverview of the decision tree technique, and how I have applied it in different ways.
Final ThoughtsLessons learned.
3
Agenda
IntroductionBackground and experience.
What is Data Mining?High level overview.
Decision Tree ExamplesOverview of the decision tree technique, and how I have applied it in different ways.
Final ThoughtsLessons learned.
4
IntroductionSanDiskJason Noriega
Senior Workforce Analyst at SanDisk
Experience: eBay, Lawrence Livermore National Laboratory, NASA, and National Geospatial Intelligence Agency.
Mobile
Computing
Consumer
5
Agenda
IntroductionBackground and experience.
What is Data Mining?High level overview.
Decision Tree ExamplesOverview of the decision tree technique, and how I have applied it in different ways.
Final ThoughtsLessons learned.
6
What is Data Mining?
Explores Your Data
Finds PatternsPerforms
Predictions
Data Mining
Targeted Customer Retention
Building Effective Marketing Campaigns
Targeted Employee Retention
Building Effective Recruiting Initiatives
Typical Uses WF Analytics
Data mining is a field that utilizes methods of machine learning,traditional statistics, database technology, and data visualization.
Predict 12 Months of Sales
of a New Product
Predict 12 Months of Employee
Separations
7
High Level Process
Predictive Model
Historical Data Predictive Model
Data with Predictions
• R Program• Weka• SAS EG• SAS JMP• SPSS Modeler
8
Agenda
IntroductionBackground and experience.
What is Data Mining?High level overview.
Decision Tree ExamplesOverview of the decision tree technique, and how I have applied it in different ways.
Final ThoughtsLessons learned.
9
Decision Tree TechniqueDecision tree algorithms applied to employee attrition data provides you with a
laser precision focus on employee retention.
Recursive Partitioning Decision Rules - Example
Predictors1. Job Function2. Gender3. Ranking4. …..
Predictors1. Age2. Rating3. Commute4. …..
Predictors1. Performance2. Tenure3. Commute
Software Engineer
Top Performer
10%
25%
35%
50%
Leavers
Leavers
Leavers
Leavers
90%Stayers
75%Stayers
65%Stayers
50%Stayers
10%Leavers
90%Stayers
25%Leavers
75%Stayers
35%Leavers
65%Stayers
50%Leavers
50%Stayers
Age < 40
Rule 1: 50% Likelihood of LeavingIf Job Function = Software EngineerAnd Age < 40And Performance = Top Performer
Rule 2: 60% Likelihood of LeavingIf Manager Effectiveness < 70% FavorableAnd Pay Grade < 17, And Region = Americas
Rule 3: 70% Likelihood of LeavingIf Not Promoted in Prior 2 Years = NoAnd Tenure > 3 Years
All data is notional
10All data is notional
Improving the Retention of Employees
Where are the high rates of voluntary
turnover?
Decision TreeAnalysis
Why do these patterns exist?
ExitSurvey
Where should we prioritize areas of
improvement?
Prioritization
What can do to reduce high rates of voluntary turnover?
Action Planning
Example 1
11
Example 1 - Variables
2010 2012
Demographics/Tenure
AgeGenderTenure
Variables for Attrition Analysis
Job ContentReasgned Diff. Pers AreaJob FunctionExempt/NonexemptLatest Perform. RatingPerform. Rating DeltaManager Flag
LocationRegionCountryLocation
CompensationPay GradeComp RatioPromotion FlagSpot Award FlagTuition Reimbursement Flag
2011
Model Timeframe for Finding Patterns
Attrition Analysis Timeframe
May
ManagerMgr TenureMgr Perform.
June
12
Example 1 - Tree# %
Stayed: 11000 86%
Left: 1800 14%
# %Stayed: 4000 80%
Left: 1000 20%
# %Stayed: 3000 86% Left: 500 14%
# %Stayed: 3000 94% Left: 250 8%
# %Stayed: 1000 95% Left: 50 5%
Employee TenureLess than 1.1 yrs 1.1-2.6 2.6-7.3 > 7.3 Yrs
# %Stayed: 2000 87%
Left: 300 13%
# %Stayed: 500 82% Left: 100 17%
# %Stayed: 1500 71%
Left: 600 29%
Pay Grade<= 16 17-22 > 22
# %Stayed: 1100 85%
Left: 200 15%
Business UnitOrganization #2Organization #1
# %Stayed: 400 50%
Left: 400 50%
# %Stayed: 100 33%
Left: 200 67%
Compa Ratio> 85<= 85
# %Stayed: 300 75%
Left: 100 25%
All data is notional
All data is notional
13
Example 1 – Survey Linkage # %Stayed: 11000 86%
Left: 1800 14%
# %Stayed: 4000 80%
Left: 1000 20%
# %Stayed: 3000 86% Left: 500 14%
# %Stayed: 3000 94% Left: 250 8%
# %Stayed: 1000 95% Left: 50 5%
Employee TenureLess than 1.1 yrs 1.1-2.6 2.6-7.3 > 7.3 Yrs
# %Stayed: 2000 87%
Left: 300 13%
# %Stayed: 500 82% Left: 100 17%
# %Stayed: 1500 71%
Left: 600 29%
Pay Grade<= 16 17-22 > 22
Exiting Employees
Job matched expectations
Opportunity for growth
High Importance/Low Performance
Low Importance/Low Performance
Low Importance/High Performance
High Importance/High Performance
All data is notional
Uncompetitive Pay
14
Example 1 - Tree
All data is notional
# %Stayed: 15000 86%
Left: 2500 14%
# %Stayed: 4000 80% Left: 1000 20%
# %Stayed: 3000 86%
Left: 500 14%
# %Stayed: 3000 90% Left: 250 10%
# %Stayed: 1000 93% Left: 50 7%
Employee TenureLess than 1.1 yrs 1.1-2.6 yrs 2.6-7.3 > 7.3 Yrs
# %Stayed: 1500 82%
Left: 200 13%
# %Stayed: 1200 93%
Left: 150 7%
# %Stayed: 300 67%
Left: 150 33%
Performance Rating
DNM, Meets Some
Exceeds SomeExceeds MostMeets
# %Stayed: 350 78%
Left: 100 22%
# %Stayed: 850 95%
Left: 50 5%
PromotionYes No
http://www.forbes.com/sites/dorieclark/2013/03/08/how-big-data-is-transforming-the-hunt-for-talent/
Big Data Forbes Article
15All data is notional
Example 2 – Grouping Variables
RecruitmentEducation LevelTotal Yrs ExperienceTotal Yrs Relevant ExperienceRecruitment Source
Location/Job/OrgPerformance RatingRegionStateCountryWork LocationPay GradeJob FunctionJob LevelBusiness Unit
Manager ImpactOverall Engagement Score
- Overall Satisfaction- Feel Proud- Feel Motivated- Recommend as Great Place
Overall Job Sat. Score- Work feels important purpose- Clearly understand indv. Objectives- Job makes good use of skills- Oppty to learn and grow
Overall Mgr Sat. Score- Mgr clearly explains expectations- Mgr provides ongoing coaching- Mgr actively supports prof. dev.- Mgr involve people with decisions- Mgr instills positive attitude- Mgr is someone I trust
Mgr PerformanceMgr Tenure
Impact of People Mgr
Location, Job,Org, Other Other
AgeGenderCompa Ratio
Where does the problem exist?
What impact doesthe mgr have?
How can we targetthose most likely to stay?
Are there any other important variables?
Other
Recruitment
16All data is notional
Where Does the Problem Exist?# %
Stayed: 1210 83%
Left: 252 17%
# %Stayed: 795 80%
Left: 194 20%
Region
Americas
# %Stayed: 83 87%
Left: 12 13%
Asia Pacific# %
Stayed: 248 93%
Left: 20 7%
EMEA
# %Stayed: 638 77%
Left: 188 23%
# %Stayed: 157 96%
Left: 6 4%
Job FunctionA, B, and C All Other
Performance Rating
# %Stayed: 84 76%
Left: 26 24%
# %Stayed: 1126 83%
Left: 226 17%
Does Not Meet
Meets Some, Meets,Exceeds Some/MostNo Rating
Location/Job/Org- Perform. Rating- Region- State- Country- Work Location- Pay Grade- Job Function- Job Level- Business Unit
17All data is notional
What Impact does the Manager Have?Mgr Impact
- Overall Engagement Score- Overall Satisfaction- Feeling Proud- Feeling Motivated- Recommend great place
- Overall Job Sat. Score- Work feels important purpose- Clearly understand indv. Obj- Job makes good use of skills- Oppty to learn and grow
- Overall Mgr Effectiveness- Mgr clearly explains expectations- Mgr provides ongoing coaching- Mgr actively supports prof. dev.- Mgr involve people with decs- Mgr instills positive attitude- Mgr is someone I trust
- Mgr Performance- Mgr Tenure # %
Stayed: 351 73%
Left: 128 27%
# %Stayed: 287 84%
Left: 60 16%
Mgr Effectiveness
<= 75% Favorable > 75% Favorable
# %Stayed: 1210 83%
Left: 252 17%
# %Stayed: 795 80%
Left: 194 20%
Region
Americas
# %Stayed: 83 87%
Left: 12 13%
Asia Pacific# %
Stayed: 248 93%
Left: 20 7%
EMEA
# %Stayed: 638 77%
Left: 188 23%
# %Stayed: 157 96%
Left: 6 4%
Job FunctionA, B, and C All Other
Performance Rating
# %Stayed: 84 76%
Left: 26 24%
# %Stayed: 1126 83%
Left: 226 17%
Does Not Meet
Meets Some, Meets,Exceeds Some/MostNo Rating
18All data is notional
Example 3 Quarterly Workforce Dashboard with Deep Dive
Predictors1. Tenure2. Age3. Region4. …..
Predictors1. Region2. Rating3. Pay Grade4. …..
Predictors1. Grade2. Performance3. Age
Tenure
<= 7
10%
25%
35%
50%
Leavers
Leavers
Leavers
Leavers
90%Stayers
75%Stayers
65%Stayers
50%Stayers
19%Leavers
81%Stayers
25%Leavers
75%Stayers
35%Leavers
65%Stayers
50%Leavers
50%Stayers
Region
Quarterly Dashboard Deep Dive
<= 2.5 yrs > 2.5 yrs
India
Grade> 7
12.3%14.2%
15.4%17.1% 19.1%
0%
5%
10%
15%
20%
25%
2013-Q2 2013-Q3 2013-Q4 2014-Q1 2014-Q2
Voluntary-ORG Involuntary -ORG Voluntary-Companywide
19
SoftwareR Program for Statistical Computing SPSS Modeler
http://www.r-project.org/
http://rattle.togaware.com/rattle-install-mswindows.html