Transcript ISM-2013-Predicting-Outcomes-in-Child-Care-LA

Data Mining Solution
Using Data Mining Analytics
to Support Fraud Detection in
CalWORKs Stage 1 Child Care
1
Data Mining Technology
Data mining is a reiterative process of selecting, exploring
and modeling large amounts of data to identify meaningful,
logical patterns and relationships among key variables.
Source:
Data Mining 101: How to Reveal New Insights in Existing Data to Improve Performance
Insights from a webinar in the SAS Applying Business Analytics Series
Originally broadcast in June 2010
2
Project Background
DPSS’ Data Mining Solution (DMS) is a computer application that employs
pattern detection and predictive analytics to detect and prevent fraud in public
assistance programs, such as our CalWORKs Stage 1 Child Care Program.

A Board Motion introduced by Supervisor Antonovich on May 29, 2007, provided the
Department of Public Social Services (DPSS) the vision to utilize cutting edge technology,
such as data mining and predictive analytics, to ensure and maintain the integrity of the
County's public assistance programs.

A successful pilot was completed in 2008 which evaluated the effectiveness of using data
mining technology to detect potential fraud in the CalWORKs Stage 1 Child Care Program.

The DMS Agreement was approved by the Board on December 22, 2009, for SAS Institute
Inc. (SAS) to design, develop and implement the data mining technology for Los Angeles
County to target fraud in the CalWORKs Stage 1 Child Care Program.

The DMS Application was implemented on May 2011 by DPSS to target fraud in the
CalWORKs Stage 1 Child Care Program.

The Board Approved an Amendment to the DMS Agreement on May 15, 2012 to extend the
data mining technology to In-Home Supportive Services (IHSS) Program.
3
Project Objectives and Key Milestones
DMS Application is hosted in Cary, North Carolina by SAS OnDemand
 The SAS Fraud Framework tracks:
• CalWORKs Stage I Child Care Participants with children requesting
assistance from the County;
• Providers that care for children while the parent or guardian go to work or
school; and
• Employers on record providing employment for the participants who attempt to
defraud the County of Los Angeles by obtaining payment for falsified services.
 Using state of the art data mining techniques the DMS application:
• Prioritizes referrals from Alternate Payment Program agencies (APPs)
• Consolidates data for investigations
• Shows networks among providers and participants
• Streamlines / optimizes searching of County data sources
• Capable of integrating with existing workflow and case management systems
 Displays results using the advanced visualization application
 Uses statistically designed risk measures to predict collusion activities
4
Project Data Sources
Data Preparation Efforts
Historical data from 2001 to Present
Data was (cleaned/matched/consolidated/geocoded) monthly from DPSS and
external data sources to generate dozens of variables for data mining models
including:
Data Focus: Participant and Provider
CalWORKs Stage 1 Child Care
–
–
–
–
Child care utilization, request and provider files
Welfare-to-Work activity tables from the GEARS system
Child care licensing files
LEADER Case and Individual tables for participant records
Data includes known cases of fraud and alleged fraud
– LEADER Fraud cluster tables to identify participants referred/prosecuted for
Child Care and other fraud types
External Data Sources
– State employment, employer and Income& Eligibility Verification System (IEVS) &
New Hire (NHR) files
– Dun and Bradstreet employment file
– In-Home Supportive Services (IHSS) participant and provider files
5
Anomaly Detection
Risk Assessment





Rules/Pattern Recognition
Anomaly Detection
Predictive Model
Hot List - (e.g., Providers & Employers with known fraudulent activity)
Social Network Linkages
Operational Outcome





Prioritized by Alert score
Monthly High Risk Report
Drill-down into case detail
Further drill-down into Alert detail
Launch into other ad-hoc analyses
Alert Score
 Base score
 Plus or minus depending on value of components
6
Utilization Process
Triage View
 High risk Alerts are generated based on the comparisons between
CalWORKs Stage 1 Child Care cases and the typical profile of fraudulent
CalWORKs cases
 Designated Triage Workers (DTWs) are assigned
comprehensive case reviews based on these Alerts
to
conduct
 Referrals are initiated to Welfare Fraud Prevention & Investigations
(WFP&I) Section
 Case action reviews result in one or more of the following outcomes:
termination of benefits, overpayments, reduction in benefits, share of cost
and/or fraud referral
Investigator View
 DMS provides tools and the capability to the DPSS WFP&I team to assist
in their detection, prevention, and investigation of fraud in the CalWORKs
Stage 1 Child Care Program
 The Social Network Analysis allows WFP&I to identify suspicious cases for
preliminary earlier investigation
 Provide access to participant’s 10-year historical data across Programs
7
LA County Fraud Framework – DMS
Log On Page
8
CASE VIEW SELECTION
Case View Select
Triage View
Triage View
Investigator View
Investigator View
User Interface Participant Details
The Participant Detail pane provides a quick view at the participant case record
information related to residential address, family members, providers, employment, source
income and benefits and prior welfare fraud historical records.
Participant Detail Page
12
Timeline Graph
The Timeline tab is a graphic representation of the data from the Provider, Address, Employment,
Component, Income and Benefits, and DPSS Actions tabs This graph provides a brief, color-coded view
of each data source, as well as a quick method for seeing when each event occurred.
Timeline Tab
Street Map View
Social Network Map View
14
Relationship Tree
The Relationship Tree tab shows a participant’s family, household members, and
other relatives. Users can adjust the time slider to view the tree over time.
Relationship Tree
Risk Assessment
Risk Assessment tab contains a list of key fraud indicators for a participant. The
tab contains two parts: the Predictive Model section and the Triggers section.
Risk Assessment
Social Network Analysis
The Social Network Analysis provides a graphical view of the Participant case record centered on
the graph with all the connections within the database of other Participants, Providers,
Employment activities and Phone Number connections to the CalWORKs Stage 1 Child Care
Participant.
Legend
Social Network Example
17
Participant Detail Summary Report in PDF Format
PDF Summary Report
18
Holistic Approach to Program Integrity
From May 2011 through July 2013, the following actions were initiated:
• * 28 Cases have been referred to the District Attorney for felony
prosecution
• * 405 DMS fraud referrals initiated for investigation
• Triage-Initiated: 311
• WFP&I-Initiated: 94
• * 753 Referrals to DPSS case workers for follow up action resulting in:
• Fraud referrals for reasons other than child care fraud
• Denial/Termination/Reduction of various public assistance benefits
• Overpayments
19