HR Analytics: Using Machine Learning to Predict Employee Turnover - Matt Dancho, Founder, Business Science
•Download as PPTX, PDF•
4 likes•2,597 views
Presented at #H2OWorld 2017 in Mountain View, CA. Enjoy the video: https://youtu.be/-qfEOwm5Th4. Learn more about H2O.ai: https://www.h2o.ai/. Follow @h2oai: https://twitter.com/h2oai. - - - In this talk, we discuss how we implemented H2O and LIME to predict and explain employee turnover on the IBM Watson HR Employee Attrition dataset. We use H2O’s new automated machine learning algorithm to improve on the accuracy of IBM Watson. We use LIME to produce feature importance and ultimately explain the black-box model produced by H2O. Matt Dancho is the founder of Business Science (www.business-science.io), a consulting firm that assists organizations in applying data science to business applications. He is the creator of R packages tidyquant and timetk and has been working with data science for business and financial analysis since 2011. Matt holds master’s degrees in business and engineering, and has extensive experience in business intelligence, data mining, time series analysis, statistics and machine learning. Connect with Matt on twitter (https://twitter.com/mdancho84) and LinkedIn (https://www.linkedin.com/in/mattdancho/).
More Related Content
What's hot
What's hot (20)
Similar to HR Analytics: Using Machine Learning to Predict Employee Turnover - Matt Dancho, Founder, Business Science
Similar to HR Analytics: Using Machine Learning to Predict Employee Turnover - Matt Dancho, Founder, Business Science (20)
More from Sri Ambati
More from Sri Ambati (20)
Recently uploaded
Recently uploaded (20)
HR Analytics: Using Machine Learning to Predict Employee Turnover - Matt Dancho, Founder, Business Science
- 2. Founder / CEO of Business Science Twitter: @bizScienc // @mdancho84 Github: business-science // mdancho84 Email: mdancho@business-science.io Website: www.business-science.io Matt Dancho
- 3. About Business Science Business & Finance Consultancy • Executive Leadership • Bolt-on Data Science Team Education & Tools • Blog: www.business-science.io • Open Source Software • Courses: Coming soon!!! Community Driven • Powered by
- 4. Adaptive Solution… 4 Learn System Model System Build ML into Decision Making ML + Leadership = Good Decision Making AI: Learning Built Into System
- 5. …Applied To Any Problem 5
- 6. Business Science: Not Just For Executives • Open Source Software • Courses (Coming Soon!) • High Demand Applications • High Demand Tools • Integrated Solutions 6 Data Scientists
- 7. Business Science: Not Just For Executives • Open Source Software • Courses (Coming Soon!) • High Demand Applications • High Demand Tools • Integrated Solutions 7 Data Scientists
- 8. Business Science: Not Just For Executives • Open Source Software • Courses (Coming Soon!) • High Demand Applications • High Demand Tools • Integrated Solutions 8 Data Scientists
- 9. Business Science: Not Just For Executives • Open Source Software • Courses (Coming Soon!) • High Demand Applications • High Demand Tools • Integrated Solutions 9 Data Scientists Learn from our blog www.business-science.io/blog
- 11. Using ML To Predict Employee Turnover HR Analytics
- 12. Three Reasons 1. Employee Attrition: A HUGE PROBLEM 2. New Techniques To Predict & Explain Turnover 3. Framework For ML In Business Applications
- 13. Three Reasons 1. Employee Attrition: A HUGE PROBLEM 2. New Techniques To Predict & Explain Turnover 3. Framework For ML In Business Applications
- 14. Three Reasons 1. Employee Attrition: A HUGE PROBLEM 2. New Techniques To Predict & Explain Turnover 3. Framework For ML In Business Applications
- 15. Three Reasons 1. Employee Attrition: A HUGE PROBLEM 2. New Techniques To Predict & Explain Turnover 3. Framework For ML In Business Applications
- 16. One More Reason R-Bloggers ● KDNuggets ● LinkedIn Code available in article: http://www.business-science.io/business/2017/09/18/hr_employee_attrition.html Just google: “Predict Employee Turnover” 4. Our Article Is Popular!
- 17. 17 Competitive Advantage “You take away our top 20 employees and overnight we become a mediocre company.” -Bill Gates
- 18. 18 Cost Of Turnover Organizations face huge costs resulting from employee turnover
- 19. 19 Cost Of Turnover Organizations face huge costs resulting from employee turnover Most important costs are intangible New Product Ideas Customer Relationships Project Management Engineering Talent
- 20. 20 Machine Learning Is Evolving • H2O • Automated Machine Learning • Predict at very high accuracy • Complex models can’t be explained • LIME • Used to explain ML classifiers • Deep learning, stacked ensembles now explainable
- 21. 21 IBM Watson HR Data Set • Simulated HR Database • Representative of real-world data • Used for IBM Watson Case Study Source: https://www.ibm.com/communities/analytics/watson-analytics-blog/hr-employee-attrition/
- 22. 22 Feature Set • HR Dataset • 35 Features • 1,470 Observations
- 23. 23 Modeling with H2O • Training the model Automated ML: -Deep Learning -Ensembles -GBM
- 24. 24 Modeling with H2O • Prediction: Test Data (Unseen) • Performance: 88% Accuracy Important for Goal Important for Business Case Puts Accuracy Into Perspective
- 25. 25 Business Implications • Recall = 62% • Will correctly classify those at risk of turnover 62 of 100 times • Critical to the business • 62% of at risk employees that can be targeted preemptively • Precision = 54% • Will avoid incorrectly assigning “Yes” 54 of 100 times • Better to target incorrectly than miss • Should not sacrifice Recall
- 26. 26 Understanding Drivers Is Key Have a great model, but… how do we prevent turnover?
- 27. 27 LIME • Local Interpretable Model-Agnostic Explanation • Theory • LIME approximates model locally as logistic or linear model • Repeats process 5000X • Outputs features that are important to local models • Result: Data Scientists Understand Why Model Predicts
- 28. 28 LIME • Complex classification models can now be interpreted • Black Box Models • Neural Networks, Ensembles, Random Forests • H2O and LIME now integrated! • https://github.com/thomasp85/lime
- 29. 29 LIME Create explainer object •Step 1: Create explainer using lime()
- 30. 30 LIME •Step 2: Create explanation using explain() Explain new observations
- 31. 31 LIME •Step 3: Plot Feature Importance
- 32. 32 LIME •Step 4: Inspect Important Features Overtime Job Role
- 33. 33 Real World: Solves Real Problems Client Case Study • Fortune 500 firm • Modeled executive potential using more sophisticated process • Our algorithm identified 16 employees that predicted as executive potential but were not targeted by client
- 34. 34 Real World: Solves Real Problems Client Case Study • Fortune 500 firm • Modeled executive potential using more sophisticated process • Our algorithm identified 16 employees that predicted as executive potential but were not targeted by client
- 35. 35 Conclusions Machine Learning Predicted Turnover • 88% Accuracy • 62% Recall Important! Machine Learning Explained Turnover • Turnover greater based on Job Role & Overtime Framework For High Accuracy & Explainability
- 36. 36 Conclusions Machine Learning Predicted Turnover • 88% Accuracy • 62% Recall Important! Machine Learning Explained Turnover • Turnover greater based on Job Role & Overtime Framework For High Accuracy & Explainability
- 37. 37 Conclusions Machine Learning Predicted Turnover • 88% Accuracy • 62% Recall Important! Machine Learning Explained Turnover • Turnover greater based on Job Role & Overtime Framework For High Accuracy & Explainability
- 38. 38 Conclusions Machine Learning Predicted Turnover • 88% Accuracy • 62% Recall Important! Machine Learning Explained Turnover • Turnover greater based on Job Role & Overtime Framework For High Accuracy & Explainability
- 39. 39 We’re Done, Right? No!! Data Science Risk How do we know model is right? • Model not back-tested Time: Cross-sectional analysis • Model not adaptive What do we do when model breaks down? • Only certainty: CHANGE
- 40. 40 We’re Done, Right? No!! Data Science Risk How do we know model is right? • Model not back-tested Time: Cross-sectional analysis • Model not adaptive What do we do when model breaks down? • Only certainty: CHANGE
- 41. 41 We’re Done, Right? No!! Data Science Risk How do we know model is right? • Model not back-tested Time: Cross-sectional analysis • Model not adaptive What do we do when model breaks down? • Only certainty: CHANGE
- 42. 42 We’re Done, Right? No!! Data Science Risk How do we know model is right? • Model not back-tested Time: Cross-sectional analysis • Model not adaptive What do we do when model breaks down? • Only certainty: CHANGE
- 43. Remember: Adaptive Solution… Learn System Model System Build ML into Decision MakingWe Build AI Into System Learning Solutions
- 44. What About Communication? 44 We Build Data Science Applications Shiny, PowerBI, Tableau www.business-science.io/demo
- 45. Twitter: @bizScienc // @mdancho84 Github: business-science // mdancho84 Email: mdancho@business-science.io Website: www.business-science.io