Predictive Analytics in Action Steven Golightly Ph.D. Director Los Angles County Child Support Services
Who Are We?
Los Angeles County LA County is 24% of the State s child support caseload. Over 1500 employees Caseload sizes averages 800 per case caseworker
Analyzing Data Program Support Division CSTATs Meeting Data Sharing
Why Do We Need It?
Business Need for Analytics Child Support Agencies Challenges Less resources staff (budget cuts, hiring freezes) money (budgets flat lined) Increased caseloads and workloads Increased performance expectations and goals
Integration of Predictive Analytics within Child Support Agencies Predictive Analytics Objectives Improved decision-making at case on-set Identify payment behavior Actuarial decisions based on statistical relationships between variables
Predictive Analytics Will Predict a cases payment behavior Caseworker will be able to: Make data-driven decisions Think strategically Reduce inefficiency Identify and confront under-performing cases Prioritize work Overturn conventional case management thinking
How We Began?
Predictive Analytics Encompasses a variety of techniques that analyze current and historical facts to make predictions about future, or otherwise unknown, events. Statistics Modeling Data mining
Model Development Procurement of Software Staffing Concerns Training of Software
Procurement of Software LA County CSSD chose IBM s SPSS Modeler Professional as our predictive analytic tool since we were already working with SPSS Statistics. Client license vs. Desktop license o Modeler could be used by a number of users from their own computers. Cost o Purchased one license, which will require users to schedule when the software can be used.
Staffing Concerns Staffing Concerns o Skill set needed for analytics o Internal staff (building up skills in analytics) o Management Fellows Program o Hiring staff with appropriate background o Unions Training of Software o Trial and error
Where Are We?
Goal Predictive Analytics Goal: By June 30, 2014, develop an initial analytical model which can be used to predict the probability that payments will be received in a child support case. Score all active enforcement cases based on those predictions and distribute them in a format that will allow Child Support officers to prioritize their work.
Scoring Project Project Task: Build an analytical model via the IBM SPSS Modeler software that will score our enforcement caseload based on our prediction that Non-Custodial Parent (NCP) will regularly pay monthly order in full.
Data Mining Identifying the right data Data Age Available Data Elements Accessibility 15% 5% 5% 5% 5% Validating Data 65% Normalizing data for software Business Understanding Data Preparation Deployment Data Understanding Modeling Evaluation
Data Management Multiple tools to manage data Microsoft Access, Excel IBM SPSS Statistics, Modeler Data transformed multiple times to get data into a format understandable to Modeler
Scoring Project Run basic statistics Identify Target Variable Sample data View data relationships (Association Model) Score data (Classification Model) IBM SPSS Modeler
Scoring Project IBM SPSS Modeler - Scores
Data Elements Scored NCP Data Age Address Cell Phone Email address Employer Information UIB, DIB Prison Indicator Bankruptcy Flag Case Information Arrears Amount Current Support Order Amount Early Intervention Participation Youngest Dependent Age
Model Analysis Model Validation Scored all enforcement cases, monthly, starting October 2012 through September 2013. Compared a years worth of predictions to actual payment history Year 1 Prediction Accuracy Payer Non-Payer Predicted Correctly 30.2% 54.8% 85% Predicted Incorrectly 6.8% 8.2% 15%
Next Steps
Upcoming Task Develop a deployment plan Regular re-evaluations of model Identify other uses of predictive analytics within department Automate modeling process
Challenges Data Clean Up Staffing Civil Service Rules Unions Areas to address: Model needs constant monitoring and modifications. Improve data accessibility. Additional training opportunities for staff. Case ownership model
Scored caseload will lead to a balanced workload for our caseworkers. Allocate cases based on case complexity (workload) vs. caseload Strategize customized action plans based on cases need Caseworkers perform efficiently; decreases collection cost and improves case performance
Thank You Dr. Steven Golightly Steven_Golightly@cssd.lacounty.gov