Early Warning System –Attrition Model using SVM
3AI September 6, 2020
Talent Analytics
Problem Statement
- Leading IT services provider, with a global workforce base of over 150k employees during that time
- High attrition across its mid-level ranks and risks losing its trained talent pool to other players in the industry
- An eclectic base of associates with varied educational backgrounds & skillsets
- Employees are spread across various business units & divergent roles
- Different business units have different rates and reasons for attrition
Analytics Led Approach
-
Data cleansing and missing value imputation
-
Check previous (two year back) model validity
-
Logistic, random forest and SVM techniques were used to choose the best prediction model to score each associate’s attrition risk
Business Impact
-
Development of monthly score cards using the model where we classify employees into buckets of Red, Amber and Green Risk attrition probabilities.
-
Those in red with favorable ratings are intervened with by talent management
Critical Success Factors
- Identify customers likely to attrite early and control unwanted talent churn
- Based on model results HR team could target 30% of the employees which covered 70% probable attrition cases
- Operational effectiveness by providing actionable insights and predictive modeling
