Monday September 12, 2011
Registration & Coffee
Multiple Case Studies: Treasury, SSA, DOD, DHS, and a Federal OIG
The High ROI of Predictive Analytics for Innovative Government Agencies
Three key ways that advanced data and text analytics can enhance the ability of government agencies to achieve their mission are: 1) streamline a key process, 2) detect and prevent fraud and improper payments, or 3) identify threats to national security. Modern organizations, public and private, are often very effective at their core tasks, so analytics usually results in an iterative, rather than transformative, improvement. Still, the impact can be dramatic.
Dr. Elder will share the story (problem, solution, and effect) of five projects conducted for some of America's most innovative government agencies.
- Text mining for quick decisions at the Social Security Administration
Eliminate Fraud/ Improper Payments:
- Tax fraud detection at the IRS
- Contract fraud detection at a Federal OIG
- Threat detection at the DoD
- Text analytics and data mining to enhance border security for DHS
Advanced Analytical Solutions in Government
Markets are changing. How can you quickly find value from the deluge of big data? The need to do more with less has never been greater. More transparent decision making, timely actions and continuous learning are vital to ongoing improvement and value creation. The leader in business analytics software and services, SAS empowers organizations to make better decisions with the data they already have. SAS helps you focus on what matters now and what will happen in the future so that you can deploy results where and when they're needed. SAS gives you THE POWER TO KNOW®.
With SAS® Analytics software you can:
- Interactively visualize, explore and communicate data discoveries.
- Transform data, including unstructured, into fact-based insights for better decisions.
- Share and deploy compelling insights for taking timely actions.
- Monitor results and facilitate continuous refinement of your analytic models.
Predictive Modeling to Detect Fraud at DFAS
The Defense Finance and Accounting Service (DFAS) uses predictive modeling to aid examiners in the detection of illegal, improper, or unusual transactions conducted against the Department of Defense (DoD) financial assets. Mr. Abbott will discuss how the DFAS was able to improve their ability to detect these payments and to reduce the manpower required to research them. By taking advantage of current data mining tools, DFAS hopes to enhance the precision of their predictions. Through a cross-validation approach and generating more than 90 models, eleven models were selected, and their individual decisions were combined via voting to make the final decision. The results improved sensitivity to the known fraud payments significantly while keeping false alarms acceptably low.
IRS Predictive Analytics Panel
The U.S. Internal Revenue Service (IRS) is recognized in the analytics and business intelligence industry for having one of the most complex yet accurate predictive analytics frameworks there is. Their focus is on finding pockets of revenue not reported or under-reported, and also looking at millions of returns quickly to see variations against constraint-based models. The IRS uses their predictive analytics tools and applications as a first step in investigating potential tax evasion, fraud, under-reporting, tax preparer noncompliance, and money-laundering. This panel of IRS experts will discuss techniques, tools, and challenges to establishing a successful predictive analytics framework in a large organization within the context of two case studies. The first case study will share how the IRS uses predictive analytics to identify workload for tax return examination. The second case study shares how the IRS is using a predictive analytics tool to build a compliance program to systematically identify certain types of return preparer noncompliance.
Curbing Crime with Predictive Analytics
Stephen Hollifield will discuss how law enforcement agencies can use business intelligence (BI) based predictive analytics to stop crime before it happens. More specifically, Hollifield will provide details on how intelligence-led policing helped the RPD become proactive rather than reactive, and reduce crime by 49 percent. The implementation of a BI and analytical solution enabled RPD to capture and analyze crime data much more proficiently, use data to appropriately deploy officers and dramatically reduce crime. Hollifield will also share how this predictive technology has helped the department save time and money.
Moving Media Industry Workforce Analytics: A Texas Case
This session will examine how predictive workforce analytics and economic models led to legislation providing incentives for the entertainment industry in Texas. Attend the session and learn: How analysts identified, classified and modeled film, television and interactive digital media workforce and consumer entertainment spending data. How data was segmented for use in creating multiple industry economic growth scenarios. How the results of predictive modeling led to legislative action.
Fighting Waste, Fraud and Abuse with DATA
Despite disagreements over the macroeconomic impact of the 2009 stimulus, the government-wide data accountability platform that the Recovery Act established is clearly a successful proof-of-concept. When accurate information about Federal spending, encoded using consistent data standards, is combined on a single, public platform – waste, fraud, and abuse are exposed. Chairman Issa's DATA Act will expand the Recovery accountability platform to all Federal spending, including grants, contracts, and agencies' own internal expenditures, and impose government-wide data standards. For the first time, inspectors general, law enforcement, watchdog groups, and the public will be able to deploy predictive analytics across the whole spectrum of Federal spending.
Identifying and Predicting Attrition in Government - The Skill Set Gap
Government cutbacks and retirement of baby boomers will yield a significant skill set gap and manpower reduction. Coupled with an overall requirement for more technology skills, governments need an understanding of what skills and experience they will lose. This session will feature a discussion of this issue along with a solution from IBM that will help governments identify what skills and employee attributes exist today and predict the effects of future attrition and retirement.
Preparing Students for the Future of Analytics
Session description forthcoming.
Text Mining and Food Safety: Using Data to Proactively Protect the Nation's Food Supply
This session covers predictive analytics as it is being used at the Food Safety Inspection Service (FSIS). The presentation will focus on how text mining enables the agency to convert large unstructured text data into structured data fields. This enables quantitative analysis that informs agency policy and decisions. The presentation will cover two specific examples and the methods used to validate the analysis.
Big Data, Crowdsourcing and Innovation in Government: With Case Studies from NASA to the NSW Traffic Authority
In an age of big and complex data and myriad analytical techniques, the governance of technical expertise is a critical issue. Yet it's rarely seriously considered. Competitions generate much needed objective information about which analysts and techniques work best in specific situations. Whereas a single data scientist can do well on a problem, how can the best one be found? And competition adds fresh eyes, new ideas and elicits greater effort (the Roger Bannister effect). Kaggle has hosted competitions that have raced to the frontier of the humanly possible in areas as diverse as prioritising preventative health care, designing games rating systems and modeling dark matter.
The presentation will include a case study of Kaggle's competition to build an algorithm which will soon enable Sydney residents to obtain probabilistic predictions of tollway travel times in just the way they are familiar with from weather forecasting.
Targeted Grant Outreach and Effective Grants Management through Predictive Analytics
COPS implemented predictive analytics in response to heightened reporting requirements from ARRA funding. This session details the agency's approach to building a system and predictive algorithm that combined two disparate datasets, financial and programmatic, across four information systems and unlocked new insights into identifying reporting anomalies, directing resources, and driving performance improvement throughout the grants management process. The session will explore the relative benefits of this investment, including process efficiencies gained, better quality data, and new methods of interpreting the data. The session also will outline the agency's vision for expanding its usage of predictive analytics moving forward.
Tuesday September 13, 2011
Registration & Coffee
Transparency and Analytics: Valuable Tools to Help Drive Accountability
Chairman Devaney will share his views on how transparency and analytics have played an important role in minimizing fraud and waste in the $787 billion dollar Recovery Program. He will also share his views on how these tools will play an increasingly bigger role going forward.
Advanced Analytics in Anomaly Detection and Electronic Discovery
This presentation will discuss the use of advanced analytics to proactively detect anomalies, particularly in the area of improper payments to stay ahead of evolving schemes and trends while driving actionable insights. It will also discuss how predictive coding and text based analytics can be used to enhance electronic discovery.
US Department of Agriculture Risk Management Agency's Crop Insurance Data Mining Program
The USDA Risk Management Agency's Crop Insurance Data Mining Program has saved American taxpayers nearly half a billion dollars by highlighting potential fraud and abuse in the program and, as a result, helped RMA to avoid making improper payments. The RMA started by building a data warehouse and integrating data on weather, soils and other argronomically relevant factors resulting in a database that contains more than 2 terabytes of information. By linking the data across time, the CAE is able to run multiyear comparisons, resulting in a key analytical approach that was previously unattainable.
Achieving Engineering Excellence through Text Mining at NASA
Databases today have become extremely sophisticated storing structured, unstructured data and documents in various formats. Legacy databases have one very large problem the percentage of inconsistent data and databases are not homogeneous, but rather heterogeneous, very distributed, require many labor hours, and very knowledgeable individuals to analyze and retrieve information. Utilizing text mining methodologies reduces labor hours within NASA's Human Space Flight Program. The results has helped NASA to evolve discovering unknown and known trends, correlation, and finding significant observations across many projects, which once was impossible by analyzing structured and unstructured data utilizing text mining.
Fighting the Good Fraud Fight: USPS OIG Contract Fraud Detection
Fraud is a costly problem for many businesses, and the efforts required to protect against it further compound the cost - and in the case of the USPS OIG – also adversely affects the Postal Service's interests. In this session, you will learn how the USPS OIG applied advanced analytics to support the Contract Fraud Program. Antonia and Bryan will show how when applying analytic tools, they are able to quantify "red flag" behavior or circumstances into suspicion scores which allow for ranking of cases to investigate or monitor.
Expert Panel: Effectively Applying Predictive Analytics in the Government
Moderator: John Elder, CEO, Elder Research, Inc., and General Chair, PAW- Gov
Predictive analytics in the government continues to become part of many of the agency missions and directives. This panel of experts will address challenges, issues and successes across a variety of government agencies and applications. Topics to be addressed include:
- How can predictive analytics be applied in a preventative way to stop fraud before it's paid, rather than the retrospective "pay and chase" model today?
- What are the emerging trends in predictive analytics applications in the government? What are the major hurdles or barriers in the adoption of predictive analytics in the government?
- Where do you see the biggest ROI (current or future) of predictive analytics within the government?
Pursuit through Cyberspace – Predictive Analytics
In this fast-changing era of globalism and the digital revolution, the way we conduct investigations and fight crimes are also changing. Data mining and predictive analytics are becoming increasingly important tools for both detection and prevention of crimes. But this migration to predictive analytics will require innovative thinking by agencies, managers, and investigators about how best to integrate these powerful new tools.
Blame it on the Nerds: Using Predictive Analytics to Optimize Postal Facilities
Everyone wants a Post Office in their neighborhood and a mail processing facility in their town. How can the Postal Service right size efficiently and fairly? Blame it on the Nerds! Learn how the USPS OIG used predictive analytics and other modeling techniques to offer strategies for optimizing the location of Post Offices and Postal Service plants.
MORSA IVA: A Model for Fraud Prevention in VAT Refunds in Mexico
The aim of the Risk Assessment Model for VAT Refunds (MORSA IVA, in spanish) is to: 1) Determine the taxpayers\' refund payment and 2) Reduce fraud regarding improper refunding. This model is built over two main pillars: the business rules that are given by the current policies and legislation, and a predictive model which identifies and assesses irregular conducts of taxpayers. MORSA IVA has been operating since 2010 and it has automated many inner processes and has given more assertive resolutions regarding tax refunds (over 12,000 tax refunds). At first, the model was intended to operate only for high-income taxpayers. However, the Risk Models Office of the Mexican Tax Administration is working in a newer version of the MORSA IVA, which will operate for the rest of the taxpayers. In this case study, we will discuss about the work we had gone through and the current challenges our people are facing while improving the model.
First Order Approximation Methodology (FOAM): What to Do When You Have 2.5 Yottabytes of Data to Analyze
The First Order Approximation Methodology (FOAM) is a statistics-based methodology for doing "first-pass" characterization of a cyber event, e.g., an intrusion. This initial approximation could be used for sensor configuration or tuning. FOAM is not new. It leverages existing statistical algorithms--Principal Component Analysis (PCA) and Logistic Regression (LR) to perform first pass approximations. This session will describe a proposal to integrate PCA and LR to perform first order predictive analytics. FOAM can be applied to any problem that attempts to: (a) reduce the dimensionality of a data set; and (b) use this reduced set for detection and classification.
Workforce Analytics: Predicting Retirement, Resignations, and More
Human resource managers perform workforce planning using a wide array of tools and intuition. Workforce planning for very large organizations can benefit from predictive models that identify employees that are likely to retire, resign, quit, be laid-off, or fired. Working with a large collection of detailed personnel and payroll data, our team applied predictive modeling techniques to build accurate models that now allow HR personnel to identify the predicted workforce status of specific individuals. Armed with these predictions, HR can identify specific skill sets that will be in short supply or over-represented, and design proactive programs to manage the workforce skill composition. The session will focus on the methods used to develop the models, the model strengths and weaknesses, with a related discussion on data preparation and enrichment.
Wednesday, September 14, 2011
The Best and the Worst of Predictive Analytics: Predictive Modeling Methods and Common Data Mining Mistakes
Click here for the detailed workshop description
Dr. Elder will describe the key inner workings of leading algorithms, demonstrate their performance with business case studies, compare their merits, and show you how to pick the method and tool best suited to each predictive analytics project. Methods covered include classical regression, decision trees, neural networks, ensemble methods, uplift modeling and more.
Workshop sponsored by:
Thursday, September 15, 2011
Hands-On Predictive Analytics with SAS Enterprise Miner
Click here for the detailed workshop description
"Hands-on Predictive Analytics" puts predictive analytics into action. This one-day workshop leads participants through the industry standard data mining process, from Business Understanding through Model Deployment, approaching each stage of this process by driving a state-of-the-art data mining software product. In this way, attendees gain direct experience applying this "best practices" process, and ramp up on an industry-leading tool to boot.