Soumettre la recherche
Mettre en ligne
1345 keynote roberts
•
11 j'aime
•
6,922 vues
Rising Media, Inc.
Suivre
P Roberts Keynote
Lire moins
Lire la suite
Données & analyses
Signaler
Partager
Signaler
Partager
1 sur 43
Recommandé
Survival Analysis for Predicting Employee Turnover
Survival Analysis for Predicting Employee Turnover
Tom Briggs
How to Start People Analytics Practice?
How to Start People Analytics Practice?
Martin Boudikianov
The Power of HR Analytics
The Power of HR Analytics
HackerEarth
The Digitization of Human Resources
The Digitization of Human Resources
Accenture Technology
People Analytics: State of the Market - Top Ten List
People Analytics: State of the Market - Top Ten List
Josh Bersin
The What, The Why and the How of People Analytics November 2017
The What, The Why and the How of People Analytics November 2017
Dave Millner
HR Analytics
HR Analytics
Shojibul Alam Shojib
Agile Recruiting White Paper
Agile Recruiting White Paper
Amber Grewal
Recommandé
Survival Analysis for Predicting Employee Turnover
Survival Analysis for Predicting Employee Turnover
Tom Briggs
How to Start People Analytics Practice?
How to Start People Analytics Practice?
Martin Boudikianov
The Power of HR Analytics
The Power of HR Analytics
HackerEarth
The Digitization of Human Resources
The Digitization of Human Resources
Accenture Technology
People Analytics: State of the Market - Top Ten List
People Analytics: State of the Market - Top Ten List
Josh Bersin
The What, The Why and the How of People Analytics November 2017
The What, The Why and the How of People Analytics November 2017
Dave Millner
HR Analytics
HR Analytics
Shojibul Alam Shojib
Agile Recruiting White Paper
Agile Recruiting White Paper
Amber Grewal
The Case for Value Stream Architecture (Mik Kersten, Carmen DeArdo)
The Case for Value Stream Architecture (Mik Kersten, Carmen DeArdo)
Carmen DeArdo
The Rise of People Analytics
The Rise of People Analytics
Human Capital Media
Predictive Analytics for HR: A Primer to Get Started on your HR Analytics Jou...
Predictive Analytics for HR: A Primer to Get Started on your HR Analytics Jou...
Dr Susan Entwisle
Agile recruiting: Optimizing your Talent Acquisition Operating Model
Agile recruiting: Optimizing your Talent Acquisition Operating Model
Beamery
People Analytics is the New HR
People Analytics is the New HR
LinkedIn Talent Solutions
How to go from structureless to structured without losing your vibe
How to go from structureless to structured without losing your vibe
Camille Fournier
Lean Project Portfolio Management
Lean Project Portfolio Management
Alexander Apostolov
Business agility
Business agility
Srinath Ramakrishnan
Agile project management
Agile project management
eng100
HR Metrics That Matter
HR Metrics That Matter
HR Tech Advisor
Agile Delivery Powerpoint Presentation Slides
Agile Delivery Powerpoint Presentation Slides
SlideTeam
The New Model for Talent Management: Agenda for 2015
The New Model for Talent Management: Agenda for 2015
Josh Bersin
People Analytics.pptx
People Analytics.pptx
Yodhia Antariksa
Top 9 business analyst interview questions answers
Top 9 business analyst interview questions answers
Jobinterviews
Hr Analytics
Hr Analytics
Atif Shaikh
Talent Acquisition Best Practices Process Map
Talent Acquisition Best Practices Process Map
blayton551
Digital HR.pptx
Digital HR.pptx
Yodhia Antariksa
RethinkingAgile_AAC2019
RethinkingAgile_AAC2019
Agile Austria Conference
Hr analytics
Hr analytics
Preksha Pagare
Value Stream Management: Is Your Organization Ready?
Value Stream Management: Is Your Organization Ready?
DevOps.com
1440 track2 roberts
1440 track2 roberts
Rising Media, Inc.
1505 Statistical Thinking course extract
1505 Statistical Thinking course extract
Jefferson Lynch
Contenu connexe
Tendances
The Case for Value Stream Architecture (Mik Kersten, Carmen DeArdo)
The Case for Value Stream Architecture (Mik Kersten, Carmen DeArdo)
Carmen DeArdo
The Rise of People Analytics
The Rise of People Analytics
Human Capital Media
Predictive Analytics for HR: A Primer to Get Started on your HR Analytics Jou...
Predictive Analytics for HR: A Primer to Get Started on your HR Analytics Jou...
Dr Susan Entwisle
Agile recruiting: Optimizing your Talent Acquisition Operating Model
Agile recruiting: Optimizing your Talent Acquisition Operating Model
Beamery
People Analytics is the New HR
People Analytics is the New HR
LinkedIn Talent Solutions
How to go from structureless to structured without losing your vibe
How to go from structureless to structured without losing your vibe
Camille Fournier
Lean Project Portfolio Management
Lean Project Portfolio Management
Alexander Apostolov
Business agility
Business agility
Srinath Ramakrishnan
Agile project management
Agile project management
eng100
HR Metrics That Matter
HR Metrics That Matter
HR Tech Advisor
Agile Delivery Powerpoint Presentation Slides
Agile Delivery Powerpoint Presentation Slides
SlideTeam
The New Model for Talent Management: Agenda for 2015
The New Model for Talent Management: Agenda for 2015
Josh Bersin
People Analytics.pptx
People Analytics.pptx
Yodhia Antariksa
Top 9 business analyst interview questions answers
Top 9 business analyst interview questions answers
Jobinterviews
Hr Analytics
Hr Analytics
Atif Shaikh
Talent Acquisition Best Practices Process Map
Talent Acquisition Best Practices Process Map
blayton551
Digital HR.pptx
Digital HR.pptx
Yodhia Antariksa
RethinkingAgile_AAC2019
RethinkingAgile_AAC2019
Agile Austria Conference
Hr analytics
Hr analytics
Preksha Pagare
Value Stream Management: Is Your Organization Ready?
Value Stream Management: Is Your Organization Ready?
DevOps.com
Tendances
(20)
The Case for Value Stream Architecture (Mik Kersten, Carmen DeArdo)
The Case for Value Stream Architecture (Mik Kersten, Carmen DeArdo)
The Rise of People Analytics
The Rise of People Analytics
Predictive Analytics for HR: A Primer to Get Started on your HR Analytics Jou...
Predictive Analytics for HR: A Primer to Get Started on your HR Analytics Jou...
Agile recruiting: Optimizing your Talent Acquisition Operating Model
Agile recruiting: Optimizing your Talent Acquisition Operating Model
People Analytics is the New HR
People Analytics is the New HR
How to go from structureless to structured without losing your vibe
How to go from structureless to structured without losing your vibe
Lean Project Portfolio Management
Lean Project Portfolio Management
Business agility
Business agility
Agile project management
Agile project management
HR Metrics That Matter
HR Metrics That Matter
Agile Delivery Powerpoint Presentation Slides
Agile Delivery Powerpoint Presentation Slides
The New Model for Talent Management: Agenda for 2015
The New Model for Talent Management: Agenda for 2015
People Analytics.pptx
People Analytics.pptx
Top 9 business analyst interview questions answers
Top 9 business analyst interview questions answers
Hr Analytics
Hr Analytics
Talent Acquisition Best Practices Process Map
Talent Acquisition Best Practices Process Map
Digital HR.pptx
Digital HR.pptx
RethinkingAgile_AAC2019
RethinkingAgile_AAC2019
Hr analytics
Hr analytics
Value Stream Management: Is Your Organization Ready?
Value Stream Management: Is Your Organization Ready?
Similaire à 1345 keynote roberts
1440 track2 roberts
1440 track2 roberts
Rising Media, Inc.
1505 Statistical Thinking course extract
1505 Statistical Thinking course extract
Jefferson Lynch
10.09.14 glassdoor webinar_all_slides
10.09.14 glassdoor webinar_all_slides
Dr. John Sullivan
Cloud: Risk profiles, Preference and Decisions in Cloud
Cloud: Risk profiles, Preference and Decisions in Cloud
Amazon Web Services
The Datafication of HR: Graduating from Metrics to Analytics
The Datafication of HR: Graduating from Metrics to Analytics
Visier
Siemens Resource Planning Case Study
Siemens Resource Planning Case Study
Greg Bailey
Ai revolution for human capital for government v2
Ai revolution for human capital for government v2
Liew Wei Da Andrew
Leveraging The Machine: The Future of People Data Is Now
Leveraging The Machine: The Future of People Data Is Now
Cornerstone OnDemand
Asking When, Not If in Predictive Modeling
Asking When, Not If in Predictive Modeling
Andrea Kropp
The Insider's Guide to Workforce Analytics
The Insider's Guide to Workforce Analytics
Visier
2017-09-Presentation (1).pptx
2017-09-Presentation (1).pptx
ssuser3fb3b5
2017-09-Presentation (1).pptx
2017-09-Presentation (1).pptx
ssuser1415bc
Scalable agile metrics to optimize flow and reduce delivery risks
Scalable agile metrics to optimize flow and reduce delivery risks
Adonis El Fakih
201505 Statistical Thinking course extract
201505 Statistical Thinking course extract
Jefferson Lynch
Complex Problem Solving and Big Data Analytics
Complex Problem Solving and Big Data Analytics
CoThink
Datascience101presentation4
Datascience101presentation4
Salford Systems
DFSS short
DFSS short
Chaitanya Chenna
9. Risk.ppt
9. Risk.ppt
Kameswara Rao Poranki
9. Risk.ppt
9. Risk.ppt
RajeevRanjan338430
9. Risk.ppt
9. Risk.ppt
HinaAsghar16
Similaire à 1345 keynote roberts
(20)
1440 track2 roberts
1440 track2 roberts
1505 Statistical Thinking course extract
1505 Statistical Thinking course extract
10.09.14 glassdoor webinar_all_slides
10.09.14 glassdoor webinar_all_slides
Cloud: Risk profiles, Preference and Decisions in Cloud
Cloud: Risk profiles, Preference and Decisions in Cloud
The Datafication of HR: Graduating from Metrics to Analytics
The Datafication of HR: Graduating from Metrics to Analytics
Siemens Resource Planning Case Study
Siemens Resource Planning Case Study
Ai revolution for human capital for government v2
Ai revolution for human capital for government v2
Leveraging The Machine: The Future of People Data Is Now
Leveraging The Machine: The Future of People Data Is Now
Asking When, Not If in Predictive Modeling
Asking When, Not If in Predictive Modeling
The Insider's Guide to Workforce Analytics
The Insider's Guide to Workforce Analytics
2017-09-Presentation (1).pptx
2017-09-Presentation (1).pptx
2017-09-Presentation (1).pptx
2017-09-Presentation (1).pptx
Scalable agile metrics to optimize flow and reduce delivery risks
Scalable agile metrics to optimize flow and reduce delivery risks
201505 Statistical Thinking course extract
201505 Statistical Thinking course extract
Complex Problem Solving and Big Data Analytics
Complex Problem Solving and Big Data Analytics
Datascience101presentation4
Datascience101presentation4
DFSS short
DFSS short
9. Risk.ppt
9. Risk.ppt
9. Risk.ppt
9. Risk.ppt
9. Risk.ppt
9. Risk.ppt
Plus de Rising Media, Inc.
1415 track 1 wu_using his laptop
1415 track 1 wu_using his laptop
Rising Media, Inc.
Matt gershoff
Matt gershoff
Rising Media, Inc.
Keynote adam greco
Keynote adam greco
Rising Media, Inc.
1620 keynote olson_using our laptop
1620 keynote olson_using our laptop
Rising Media, Inc.
1530 track 2 stuart_using our laptop
1530 track 2 stuart_using our laptop
Rising Media, Inc.
1530 track 1 fader_using our laptop
1530 track 1 fader_using our laptop
Rising Media, Inc.
1415 track 2 richardson
1415 track 2 richardson
Rising Media, Inc.
1215 daa lunch owusu_using our laptop
1215 daa lunch owusu_using our laptop
Rising Media, Inc.
1215 daa lunch a bos intro slides_using our laptop
1215 daa lunch a bos intro slides_using our laptop
Rising Media, Inc.
915 e metrics_claudia perlich
915 e metrics_claudia perlich
Rising Media, Inc.
855 sponsor movassate_using our laptop
855 sponsor movassate_using our laptop
Rising Media, Inc.
1615 plack using our laptop
1615 plack using our laptop
Rising Media, Inc.
1530 rimmele do not share
1530 rimmele do not share
Rising Media, Inc.
1325 keynote yale_pdf shareable
1325 keynote yale_pdf shareable
Rising Media, Inc.
1115 fiztgerald schuchardt
1115 fiztgerald schuchardt
Rising Media, Inc.
1000 kondic do not share
1000 kondic do not share
Rising Media, Inc.
905 keynote peele_using our laptop
905 keynote peele_using our laptop
Rising Media, Inc.
Stephen morse sharable
Stephen morse sharable
Rising Media, Inc.
Elder shareable
Elder shareable
Rising Media, Inc.
1115 ramirez using our laptop
1115 ramirez using our laptop
Rising Media, Inc.
Plus de Rising Media, Inc.
(20)
1415 track 1 wu_using his laptop
1415 track 1 wu_using his laptop
Matt gershoff
Matt gershoff
Keynote adam greco
Keynote adam greco
1620 keynote olson_using our laptop
1620 keynote olson_using our laptop
1530 track 2 stuart_using our laptop
1530 track 2 stuart_using our laptop
1530 track 1 fader_using our laptop
1530 track 1 fader_using our laptop
1415 track 2 richardson
1415 track 2 richardson
1215 daa lunch owusu_using our laptop
1215 daa lunch owusu_using our laptop
1215 daa lunch a bos intro slides_using our laptop
1215 daa lunch a bos intro slides_using our laptop
915 e metrics_claudia perlich
915 e metrics_claudia perlich
855 sponsor movassate_using our laptop
855 sponsor movassate_using our laptop
1615 plack using our laptop
1615 plack using our laptop
1530 rimmele do not share
1530 rimmele do not share
1325 keynote yale_pdf shareable
1325 keynote yale_pdf shareable
1115 fiztgerald schuchardt
1115 fiztgerald schuchardt
1000 kondic do not share
1000 kondic do not share
905 keynote peele_using our laptop
905 keynote peele_using our laptop
Stephen morse sharable
Stephen morse sharable
Elder shareable
Elder shareable
1115 ramirez using our laptop
1115 ramirez using our laptop
Dernier
Sealdah % High Class Call Girls Kolkata - 450+ Call Girl Cash Payment 8005736...
Sealdah % High Class Call Girls Kolkata - 450+ Call Girl Cash Payment 8005736...
HyderabadDolls
5CL-ADBA,5cladba, Chinese supplier, safety is guaranteed
5CL-ADBA,5cladba, Chinese supplier, safety is guaranteed
amy56318795
TrafficWave Generator Will Instantly drive targeted and engaging traffic back...
TrafficWave Generator Will Instantly drive targeted and engaging traffic back...
SOFTTECHHUB
Vadodara 💋 Call Girl 7737669865 Call Girls in Vadodara Escort service book now
Vadodara 💋 Call Girl 7737669865 Call Girls in Vadodara Escort service book now
gargpaaro
Gulbai Tekra * Cheap Call Girls In Ahmedabad Phone No 8005736733 Elite Escort...
Gulbai Tekra * Cheap Call Girls In Ahmedabad Phone No 8005736733 Elite Escort...
gragchanchal546
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
nirzagarg
Jodhpur Park | Call Girls in Kolkata Phone No 8005736733 Elite Escort Service...
Jodhpur Park | Call Girls in Kolkata Phone No 8005736733 Elite Escort Service...
HyderabadDolls
RESEARCH-FINAL-DEFENSE-PPT-TEMPLATE.pptx
RESEARCH-FINAL-DEFENSE-PPT-TEMPLATE.pptx
ronsairoathenadugay
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Valters Lauzums
Digital Transformation Playbook by Graham Ware
Digital Transformation Playbook by Graham Ware
Graham Ware
Abortion pills in Jeddah | +966572737505 | Get Cytotec
Abortion pills in Jeddah | +966572737505 | Get Cytotec
Abortion pills in Riyadh +966572737505 get cytotec
Reconciling Conflicting Data Curation Actions: Transparency Through Argument...
Reconciling Conflicting Data Curation Actions: Transparency Through Argument...
Bertram Ludäscher
Ranking and Scoring Exercises for Research
Ranking and Scoring Exercises for Research
Rajesh Mondal
Charbagh + Female Escorts Service in Lucknow | Starting ₹,5K To @25k with A/C...
Charbagh + Female Escorts Service in Lucknow | Starting ₹,5K To @25k with A/C...
HyderabadDolls
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
nirzagarg
20240412-SmartCityIndex-2024-Full-Report.pdf
20240412-SmartCityIndex-2024-Full-Report.pdf
khraisr
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
gajnagarg
Dubai Call Girls Peeing O525547819 Call Girls Dubai
Dubai Call Girls Peeing O525547819 Call Girls Dubai
kojalkojal131
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
gajnagarg
Top profile Call Girls In Latur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Latur [ 7014168258 ] Call Me For Genuine Models We ...
gajnagarg
Dernier
(20)
Sealdah % High Class Call Girls Kolkata - 450+ Call Girl Cash Payment 8005736...
Sealdah % High Class Call Girls Kolkata - 450+ Call Girl Cash Payment 8005736...
5CL-ADBA,5cladba, Chinese supplier, safety is guaranteed
5CL-ADBA,5cladba, Chinese supplier, safety is guaranteed
TrafficWave Generator Will Instantly drive targeted and engaging traffic back...
TrafficWave Generator Will Instantly drive targeted and engaging traffic back...
Vadodara 💋 Call Girl 7737669865 Call Girls in Vadodara Escort service book now
Vadodara 💋 Call Girl 7737669865 Call Girls in Vadodara Escort service book now
Gulbai Tekra * Cheap Call Girls In Ahmedabad Phone No 8005736733 Elite Escort...
Gulbai Tekra * Cheap Call Girls In Ahmedabad Phone No 8005736733 Elite Escort...
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
Top profile Call Girls In Bihar Sharif [ 7014168258 ] Call Me For Genuine Mod...
Jodhpur Park | Call Girls in Kolkata Phone No 8005736733 Elite Escort Service...
Jodhpur Park | Call Girls in Kolkata Phone No 8005736733 Elite Escort Service...
RESEARCH-FINAL-DEFENSE-PPT-TEMPLATE.pptx
RESEARCH-FINAL-DEFENSE-PPT-TEMPLATE.pptx
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Digital Transformation Playbook by Graham Ware
Digital Transformation Playbook by Graham Ware
Abortion pills in Jeddah | +966572737505 | Get Cytotec
Abortion pills in Jeddah | +966572737505 | Get Cytotec
Reconciling Conflicting Data Curation Actions: Transparency Through Argument...
Reconciling Conflicting Data Curation Actions: Transparency Through Argument...
Ranking and Scoring Exercises for Research
Ranking and Scoring Exercises for Research
Charbagh + Female Escorts Service in Lucknow | Starting ₹,5K To @25k with A/C...
Charbagh + Female Escorts Service in Lucknow | Starting ₹,5K To @25k with A/C...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Tumkur [ 7014168258 ] Call Me For Genuine Models We...
20240412-SmartCityIndex-2024-Full-Report.pdf
20240412-SmartCityIndex-2024-Full-Report.pdf
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
Top profile Call Girls In bhavnagar [ 7014168258 ] Call Me For Genuine Models...
Dubai Call Girls Peeing O525547819 Call Girls Dubai
Dubai Call Girls Peeing O525547819 Call Girls Dubai
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
Top profile Call Girls In dimapur [ 7014168258 ] Call Me For Genuine Models W...
Top profile Call Girls In Latur [ 7014168258 ] Call Me For Genuine Models We ...
Top profile Call Girls In Latur [ 7014168258 ] Call Me For Genuine Models We ...
1345 keynote roberts
1.
How to Use
Survival Analytics to Predict Employee Turnover (Or, Why You Shouldn't Use Logistic Regression to Predict Attrition) Talent Analytics, Corp. Pasha Roberts, Chief Scientist May 16, 2017 San Francisco, CA USA Demo Code: https://github.com/talentanalytics/class_survival_101/ www.talentanalytics.com © 2017 Talent Analytics, Corp 1 / 41
2.
▸ ▸ ▹ ▹ ▸ ▸ Who is Talent
Analytics? Predictive Modeling Platform – Advisor™ We predict employee attrition and performance pre-hire. Much like credit risk modeling: Predict likelihood to pay / default on mortgage, before extending credit Predict likelihood to perform / leave role or company early, before extending job offer PAAS (Prediction As A Service) Seamless deployment of predictions into talent acquisition process www.talentanalytics.com © 2017 Talent Analytics, Corp 2 / 41
3.
Turnover and Tenure: A
Timey-Wimey Relationship www.talentanalytics.com © 2017 Talent Analytics, Corp 3 / 41
4.
▸ ▸ Two Different (But
Related) Things Turnover: Percent of employees that terminate within a period of time Tenure: Employees’ length of time working at a role or company Is it a Wave or a Particle? www.talentanalytics.com © 2017 Talent Analytics, Corp 4 / 41
5.
Low Tenure, High
Turnover www.talentanalytics.com © 2017 Talent Analytics, Corp 5 / 41
6.
High Tenure, Low
Turnover www.talentanalytics.com © 2017 Talent Analytics, Corp 6 / 41
7.
High Tenure, High
Turnover (in 2015) www.talentanalytics.com © 2017 Talent Analytics, Corp 7 / 41
8.
Low Tenure, Low
Turnover (in Nov, Dec) www.talentanalytics.com © 2017 Talent Analytics, Corp 8 / 41
9.
▸ ▸ ▸ Main Problem: We
Can't See the Future Everyone will terminate. Someday, somehow... But when? Technical Lingo: Right Censoring www.talentanalytics.com © 2017 Talent Analytics, Corp 9 / 41
10.
▸ ▸ ▸ Each Side Has
Missing Information Turnover Has No Information About Tenure Treats a temp the same as a seasoned veteran Ignores patterns - such as people quitting right after a bonus Tenure Has No Information About Turnover High Tenure = staying power or retirement risk? Each Side Varies Depending on When You Look www.talentanalytics.com © 2017 Talent Analytics, Corp 10 / 41
11.
▸ ▸ Annual Attrition Timing
is Arbitrary What if you could see attrition at every point in time? “One Year” only matters to accountants and astronomers Cumulative Breakeven matters more to the business www.talentanalytics.com © 2017 Talent Analytics, Corp 11 / 41
12.
Survival 101 www.talentanalytics.com ©
2017 Talent Analytics, Corp 12 / 41
13.
▸ ▸ ▸ ▸ ▸ ▸ Survival Analytics Medical Roots
- How long will a patient survive a disease? Engineering Roots - How long until a machine fails? Social Science - How long do people live? Employment - How long until an employee terminates? or promotes? In General - “Expected duration of time until one or more events happen” Artfully combines turnover and tenure Lots of competing jargon and syntax from many application domains: Failure Rate, Hazard Rate, Force of Mortality, ... www.talentanalytics.com © 2017 Talent Analytics, Corp 13 / 41
14.
▸ ▸ ▸ ▸ Hazard Rate: Conditional Daily
Risk of Termination Every day in tenure there is a (small) probability that the employee will terminate Conditional on survival to the prior day Like an actuarial life table Can rise at different times: e.g. training, review or bonus time www.talentanalytics.com © 2017 Talent Analytics, Corp 14 / 41
15.
▸ ▸ ▸ ▸ Survival Rate: Daily
Probability of Being Employed at Time t How likely are you to be working here, t days after being hired? Near 100% on Day 1 (some people actually never show up) Downhill from there - usually not linear One role at a time www.talentanalytics.com © 2017 Talent Analytics, Corp 15 / 41
16.
▸ ▸ ▸ ▸ ▸ ▸ ▹ ▹ Survival Rate: Daily
Probability of Being Employed at Time t How likely are you to be working here, t days after being hired? Near 100% on Day 1 (some people actually never show up) Downhill from there - usually not linear One role at a time This is the full picture Simple transformation of the Hazard Curve Best to use statistical tool (R, SAS) Apparently possible in Excel www.talentanalytics.com © 2017 Talent Analytics, Corp 15 / 41
17.
Attrition is Built
Into Survival Curves www.talentanalytics.com © 2017 Talent Analytics, Corp 16 / 41
18.
Shelf Pattern Sparse
Pattern Many Shapes www.talentanalytics.com © 2017 Talent Analytics, Corp 17 / 41
19.
Heavy Terms During
Training Heavy Terms After 1-Year Bonus Cliffs or Kinks in the Survival Curve www.talentanalytics.com © 2017 Talent Analytics, Corp 18 / 41
20.
Everyone Has a
Survival Curve www.talentanalytics.com © 2017 Talent Analytics, Corp 19 / 41
21.
Predicting Survival To the
Individual Employee www.talentanalytics.com © 2017 Talent Analytics, Corp 20 / 41
22.
▸ ▸ ▸ ▹ ▸ ▸ ▸ Don't Use Logistic
Methods to Predict Attrition! Commonly Done: Predict one-year attrition with logistic Use tenure as a variable along with others Common Result: “The biggest cause of termination is tenure” Not an actionable coef cient Ignores nuances available to survival methods Mishandles current employee tenure Less accurate www.talentanalytics.com © 2017 Talent Analytics, Corp 21 / 41
23.
▸ ▸ ▸ ▸ Proportional Hazard Most Survival
models are “Proportional Hazard” Start with Base Survival rate Convert to Base Cumulative Hazard rate Multiply Cumulative Hazard by a single factor (hence “Proportional Hazard") Linear changes to Hazard lead to a rotation of Survival Curve Predictive Goal: Predict the amount of Survival Curve rotation for each candidate www.talentanalytics.com © 2017 Talent Analytics, Corp 22 / 41
24.
▸ ▹ ▹ ▹ ▹ ▸ ▸ ▸ ▸ Proportional Hazard Prediction
Methods Predicting a Single Number Cox regression Vanilla Cox Stepwise ElasticNet or LASSO Variable selection with Random Forest Decision Trees Random Forests Neural Networks SVM, etc There are also other survival algorithms, including parametric methods www.talentanalytics.com © 2017 Talent Analytics, Corp 23 / 41
25.
▸ ▸ ▸ ▸ Build/Validate a Survival
Model With Thresholds The Model Predicts: “Dark Blue” hires will survive the longest “Light Blue” hires will survive above-median “Grey” survival curve is current median survival. “Light Red” and “Dark Red” candidates are not likely to survive. www.talentanalytics.com © 2017 Talent Analytics, Corp 24 / 41
26.
Deeper Dive: Predicting Survival with
Proportional Hazard Follow Demo Code on GitHub: https://github.com/talentanalytics/class_survival_101/ www.talentanalytics.com © 2017 Talent Analytics, Corp 25 / 41
27.
▸ ▸ ▸ ▸ ▸ ▸ Formulas and Jargon Hazard
Function: Probability of termination at time t, conditional on survival up to time t Ranges from 0 to 1 (typically very small) Cumulative Hazard Function: Cumulative conditional probability of termination up to time t Ranges from 0 to 4ish (theoretically to In nity) Survival Function: Probability that termination will be later than time t Ranges from 1 to 0 Theoretically the formulas above are continuous measures, but in practice are discrete daily. www.talentanalytics.com © 2017 Talent Analytics, Corp h(t) H(t) = h(x) dx = −log(S(t))∫ t 0 S(t) = e −H(t) 26 / 41
28.
Calculating Survival in
R Four Simple Steps with R survival package: 1. Gather and Prepare the Data 2. Calculate Baseline Survival Function 3. Model Proportional Hazard with Cox Regression 4. Validate Model www.talentanalytics.com © 2017 Talent Analytics, Corp (t)S0 27 / 41
29.
▸ ▸ ▸ Calculating Survival in
R Four Simple Steps with R survival package: 1. Gather and Prepare the Data 2. Calculate Baseline Survival Function 3. Model Proportional Hazard with Cox Regression 4. Validate Model 5. Deploy Model into hiring 6. Hire people likely to stay on the job longer 7. Save and earn money Avoid attributing savings to soft factors like ripple effect, morale, engagement Hard savings from reduced Replacement Cost Hard earnings from increased Employee Lifetime Value www.talentanalytics.com © 2017 Talent Analytics, Corp (t)S0 27 / 41
30.
▸ ▹ ▸ ▹ ▹ 1) Gather Employee
Data > attr.data emp.id hire.date term.date factor.x factor.y factor.z <int> <dttm> <dttm> <dbl> <dbl> <dbl> 1 52 2016-02-23 <NA> 0.09555369 1.515028 102.3811 2 261 2014-10-30 <NA> 2.40144024 9.065855 100.5727 3 193 2013-12-25 2015-03-28 1.88463382 4.331642 105.6887 4 150 2013-03-27 2013-09-19 3.81625316 7.363454 110.6842 5 31 2015-03-25 <NA> 5.17392728 11.824601 110.0590 Simple Requirements Just hire.date and term.date is all you need from HRIS Leave term.date as NA if the employee haven't terminated yet Merge with predictive input (independent) variables e.g. assessment results, experience, CV detail, social media, semantic Here we use factor.x, factor.y and factor.z www.talentanalytics.com © 2017 Talent Analytics, Corp 28 / 41
31.
▸ ▸ ▹ ▹ ▸ ▹ Calculate Fields for
Survival > attr.data %>% dplyr::select(-dplyr::contains("factor")) emp.id hire.date term.date is.term end.date tenure.years <int> <dttm> <dttm> <lgl> <dttm> <dbl> 1 52 2016-02-23 <NA> FALSE 2017-05-01 1.1854894 2 261 2014-10-30 <NA> FALSE 2017-05-01 2.5023956 3 193 2013-12-25 2015-03-28 TRUE 2015-03-28 1.2539357 4 150 2013-03-27 2013-09-19 TRUE 2013-09-19 0.4818617 5 31 2015-03-25 <NA> FALSE 2017-05-01 2.1026694 Simple Derivations is.term: Have they terminated? Is term.date NA? end.date: Either term.date, or censor date, depending on is.term Could put transfer.date here with no is.term The last date we know the employee was at work tenure.years: What is employee tenure from hire.date to end.date? All we know about the employee's tenure, as of censor date www.talentanalytics.com © 2017 Talent Analytics, Corp 29 / 41
32.
▸ ▸ ▸ Dataset is Ready >
glimpse(attr.data) Observations: 400 Variables: 14 $ label <fctr> a, a, a, a, a, a, a, a, a, a, a, a, a, a, a, a, a, a,... $ hire.date <dttm> 2013-04-20, 2016-01-14, 2013-09-20, 2014-05-17, 2016-... $ term.date <dttm> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 2016-... $ is.term <lgl> FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE... $ end.date <dttm> 2017-05-01, 2017-05-01, 2017-05-01, 2017-05-01, 2017-... $ tenure.years <dbl> 4.02975576, 1.29469635, 3.61029191, 2.95626651, 1.1463... $ factor.x <dbl> 7.44846858, 1.08376058, 2.81983987, 0.56880027, 0.7868... $ factor.y <dbl> 16.838313, 9.335225, 9.519951, 12.697147, 12.912898, 1... $ factor.z <dbl> 103.10732, 100.37171, 99.76382, 95.88645, 97.61509, 10... $ emp.id <int> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,... $ scale.x <dbl> 2.41351313, -0.24754913, 0.47829958, -0.46285224, -0.3... $ scale.y <dbl> 1.31773198, -0.57718839, -0.53053544, 0.27187182, 0.32... $ scale.z <dbl> 0.3407242, -0.1849069, -0.3017096, -1.0467221, -0.7145... $ is.training <lgl> TRUE, TRUE, TRUE, FALSE, FALSE, FALSE, TRUE, TRUE, TRU... Example is 400-row simulated dataset Scaled the input factors to consistent z-scale Separated into 80% training and 20% validation datasets www.talentanalytics.com © 2017 Talent Analytics, Corp 30 / 41
33.
▸ 2) Calculate Baseline
Survival ( ) survival::Surv is a single object to hold Tenure and Turnover at once > surv.obj <- survival::Surv(training.data$tenure.years, training.data$is.term) Needs “time variable” tenure.years and “event variable” is.term survival::survfit is the Kaplan-Meier Estimator > surv.fit <- survival::survfit(surv.obj ~ 1) > summary(surv.fit) Call: survfit(formula = surv.obj ~ 1) time n.risk n.event survival std.err lower 95% CI upper 95% CI 0.000 320 1 0.997 0.00312 0.991 1.000 0.249 308 1 0.994 0.00448 0.985 1.000 0.298 303 1 0.990 0.00554 0.980 1.000 0.320 298 1 0.987 0.00644 0.974 1.000 www.talentanalytics.com © 2017 Talent Analytics, Corp (t)S0 31 / 41
34.
Native Survival Plot
Crafted ggplot2 Plot Plot Baseline Survival www.talentanalytics.com © 2017 Talent Analytics, Corp 32 / 41
35.
▸ ▸ ▸ ▸ ▸ 3) Back to
Proportional Hazard Consider a multiplier x that centers on 1 Multiply for proportional (linear) changes to Hazard Survival Curve rotates in response: Beware - inverse relationship: x ← 1.1 will increase , decrease x ← 0.9 will decrease , increase So we just need to predict x to predict Survival One number drives the entire curve, based on baseline www.talentanalytics.com © 2017 Talent Analytics, Corp (t) = (t)xHx H0 (t) =Sx e − (t)xH0 (t)Hx (t)Sx (t)Hx (t)Sx (t)H0 33 / 41
36.
Model Proportional Hazard with
Cox Regression > cox.model <- survival::coxph(formula = surv.obj ~ scale.x + scale.y + scale.z, data = training.data) > cox.model Call: survival::coxph(formula = surv.obj ~ scale.x + scale.y + scale.z, data = training.data) coef exp(coef) se(coef) z p scale.x -0.4298 0.6506 0.0918 -4.68 0.0000028 scale.y -0.3204 0.7258 0.0957 -3.35 0.00082 scale.z -0.1967 0.8215 0.0984 -2.00 0.04558 Likelihood ratio test=38.2 on 3 df, p=0.000000025 n= 320, number of events= 125 www.talentanalytics.com © 2017 Talent Analytics, Corp 34 / 41
37.
▸ ▹ ▹ ▸ ▹ ▹ 4) Predict validation.data with
New Cox Model > cox.pred <- predict(cox.model, newdata = validation.data, type = "lp") Note: we built model with training.data, now we predict against validation.data Our model has never seen anything in validation.data Great test to see how closely our predictions match known outcomes The predict.coxph() function returns 5 different types of predictions We want type = "lp" which is log(multiplier) to baseline Other types won't work with survivalROC() www.talentanalytics.com © 2017 Talent Analytics, Corp (t)H0 35 / 41
38.
▸ ▸ ▸ ▸ How does ROC
Work with Survival? > roc.obj <- survivalROC::survivalROC(Stime = validation.data$tenure.years, status = validation.data$is.term, marker = cox.pred, predict.time = 1, lambda = 0.003) > roc.obj$AUC [1] 0.7102442 ROC typically evaluates classification methods So, we make this a classi cation problem Classify: Was the employee terminated at t = 1 or not? Compare actual vs. predicted classi cations at all cutpoints Our AUC is 0.71 Not bad for a censored attrition prediction! www.talentanalytics.com © 2017 Talent Analytics, Corp 36 / 41
39.
Plot AUC with
ggplot2 www.talentanalytics.com © 2017 Talent Analytics, Corp 37 / 41
40.
▸ ▸ ▸ ▸ Don't Get Fooled
by Randomness We are predicting AUC, too Let's try predicting 1000 random samples of the same underlying pattern > test.repl <- replicate(1000, demoPrediction(verbose = FALSE)) > summary(test.repl) Min. 1st Qu. Median Mean 3rd Qu. Max. 0.3470 0.6191 0.6833 0.6771 0.7356 0.9446 AUC generally fell between 0.619 and 0.735, but 50% were outside this range The likely AUC is closer to 0.68 - still good What if you got a lucky (or unlucky) 80/20 validation split? www.talentanalytics.com © 2017 Talent Analytics, Corp 38 / 41
41.
▸ ▸ Multiple Cross-Validation Raises the
Bar Against Getting Fooled Typically run 20-100 CV iterations Can consume lots of core-hours if nested for hyperparameters www.talentanalytics.com © 2017 Talent Analytics, Corp 39 / 41
42.
▸ ▸ ▸ ▸ ▸ Download and Experiment
With this Code These slides will be on PAW website Download demo code and doc from https://github.com/talentanalytics/class_survival_101/ Clone it, fork it, run it Send bug xes Let me know how it goes www.talentanalytics.com © 2017 Talent Analytics, Corp 40 / 41
43.
Questions and Discussion Pasha
Roberts pasha@talentanalytics.com Follow Demo Code on GitHub: https://github.com/talentanalytics/class_survival_101/ www.talentanalytics.com © 2017 Talent Analytics, Corp 41 / 41