ABUSE DEFINITION BUILD-UP PRINCIPLES

Transcription

1 CONSTRUCTING SCORING METHODS FOR DETECTING QUESTIONABLE CLAIMS FOR AUTO AND HOMEOWNERS INSURANCE Richard A. Derrig, Ph.D. President. OPAL Consulting LLC Visiting Professor, Temple University CAS Spring Meeting May 17, 2010 Palm Beach, Florida Agenda Questionable Claims What Is Fraud? Hard v Soft Motivation for Scoring: Auto BI Supervised versus Unsupervised Two Unsupervised Methods: SOM Feature Maps; PRIDIT Questions and Comments Fraud Definition PRINCIPLES Clear and willful act Proscribed by law Obtaining money or value Under false pretenses Abuse: Fails one or more Principles

2 ABUSE DEFINITION PRINCIPLES BUILD-UP Not (Criminal) Fraud Unwanted, Unintended, Unnecessary Claims Disputable Damages Civil Matter Company s Problem Regulator s Problem

3 Derrig Top Five Fraud Ideas 1. FRAUD is ambiguous, ill-defined. 2. FRAUD should be reserved for criminal behavior (Hard Fraud). Abuse (Soft Fraud) 3. FRAUD ambiguity muddles the discussion and responsibility. Criminal Justice v Claim Management. Both are necessary (CIFI) 4. Criminal Fraud is several orders of magnitude less than popular estimates. 5. Fraud and Systematic Abuse can and should be mitigated by computer-assisted trained adjusters finding Questionable Claims and special investigators. Fraud Types Insurer Fraud Fraudulent Company Fraudulent Management Agent Fraud No Policy False Premium Company Fraud Embezzlement Inside/Outside Arrangements Claim Fraud Claimant/Insured Providers/Rings Non-Criminal Fraud General Deterrence Ins. System, Ins Dept + DIA, Medical & Bar Associations, o s, Other Government e Oversight, Fraud Bureaus (CIFI in Mass.) Specific Deterrence Company SIU, Auditor, Data, Predictive Modeling for Claims and Underwriting.

4 10% HOW MUCH CLAIM FRAUD? Fraud COSTS Fraud (Criminal, Hard) Small Mass. Auto & WC < 1% Abuse (Not Criminal, Soft Fraud) BIG Bucks, Depends on Line Abuse is (legally) a gray area, unethical behavior Abuse Containment is a Matter for Company/Industry/Regulator

5 TYPES OF FRAUD AUTO INSURANCE Bodily Injury -Staged Accidents -Actual Accidents/Faked Injuries -Jump-Ins -Provider Abuse / False Billing Vehicle Damage -Staged Thefts -Chop Shops -Body Shop Fraud -Adjuster Fraud -Burying The Deductible Myth

6 Major Kinds of Predictive Modeling Supervised learning Most common situation A dependent variable Frequency Loss ratio Fraud/no fraud Some methods Regression CART Some neural networks MARS Logistic Regression Naïve Bayes Unsupervised learning No dependent variable Group like records together A group of claims with similar characteristics might be more likely to be fraudulent Ex: Territory assignment, Text Mining Some methods Association rules K-means clustering Kohonen neural networks PRIDIT DATA Computers advance

7 DM Databases Scoring Functions Graded Output Using Kohonen s Self-Organizing Feature Map to Uncover Automobile Bodily Injury Claims Fraud PATRICK L. BROCKETT Gus S. Wortham Chaired Prof. of Risk Management University of Texas at Austin XIAOHUA XIA University of Texas, at Austin RICHARD A. DERRIG Senior Vice President Automobile Insurers Bureau of Massachusetts Vice President of Research Insurance Fraud Bureau of Massachusetts MAPPING: PATTERNS-TO-UNITS ns Pattern Non-Suspicious Claims Routine Claims Suspicious Claims Complicated Claims JOURNAL OF RISK AND INSURANCE, 65:2, , 274, 1998,

8 FEATURE MAP SUSPICION LEVELS S16 S S S7 1-2 S4 0-1 S1 FEATURE MAP SIMILIARITY OF A CLAIM S16 S13 S10 S7 S4 S DATA MODELING EXAMPLE: CLUSTERING Data on 16,000 Medicaid providers analyzed by unsupervised neural net Neural network clustered Medicaid providers based on 100+ features Investigators validated a small set of known fraudulent providers Visualization tool displays clustering, showing known fraud and abuse Subset of 100 providers with similar patterns investigated: Hit rate > 70% Cube size proportional to annual Medicaid revenues 1999 Intelligent Technologies Corporation

9 Fraud Classification Using Principal Component Analysis of RIDITs PATRICK L. BROCKETT Gus S. Wortham Chaired Prof. of Risk Management University of Texas at Austin RICHARD A. DERRIG Senior Vice President Automobile Insurers Bureau of Massachusetts Vice President of Research Insurance Fraud Bureau of Massachusetts LINDA L. GOLDEN Marlene & Morton Meyerson Centennial Professor in Business University of Texas Austin, Texas ARNOLD LEVINE Professor Emeritus Department of Mathematics Tulane University New Orleans LA MARK ALPERT Professor of Marketing University of Texas Austin, Texas JOURNAL OF RISK AND INSURANCE, 69:3, SEPT THIS CLASSIFICATION PROBLEM Data: Features have no natural metric-scale Target: Quality, not Quantity, Latent Variable Model: Stochastic process may have no parametric form Classification: Inverse image of one dimensional scoring function and decision rule, quality order Feature Value: Identify which features are important

10 Problem: Categorical Variables It is not clear how to best perform Principal Components/Factor Analysis on categorical variables. For PRIDIT, The categories may be coded as a series of binary (dummy) variables If the categories are ordered categories, you may apply PRIDIT and gain relationships This is the problem that PRIDIT addresses PRIDIT METHOD OVERVIEW 1. DATA: N Claims, T Features, K sub T Ranked Responses, Monotone In a Latent Variable such as Fraud 2. RIDIT score each possible response: proportion below minus proportion above, score centered at zero. 3. RESPONSE WEIGHTS: Principal Component of Claims x Features with RIDIT in Cells. 4. SCORE: Sum weights x claim RIDIT score is unique. 5. PARTITION: above and below zero is unique. RIDIT Variables are ordered so that lowest value is associated with highest or lowest probability of the latent variable, for example fraud Use Cumulative distribution of claims at each value, i, to create RIDIT statistic for claim t, value i

11 Application to Questionable Claim Detection Hospital quality was evenly distributed Variable Example Treatment Variables Computation of PRIDIT Scores Variable Label Proportion of B t1 Yes ("Yes") B t2 ( No") Assign a score to each claim The score orders the claims in an increasing estimate of the fraud or abuse content The score can be used to sort claims More effort expended on questionable claims Those more likely to be fraudulent or abusive Become the high priority for scrutiny Large # of Visits to Chiropractor Chiropractor provided 3 or more modalities on most visits Large # of visits to a physical therapist MRI or CT scan but no inpatient hospital charges Use of high volume medical provider Significant gaps in course of treatment Treatment was unusually prolonged (> 6 months) Indep. Medical examiner questioned extent of treatment Lots of hospitals in the middle, a few outliers of high and low quality Medical audit raised questions about charges 9

12 Variable Large # of Visits to Chiropractor Chiropractor provided 3 or more modalities on most visits Large # of visits to a physical therapist MRI or CT scan but no inpatient hospital charges Use of high volume medical provider Significant gaps in course of treatment TABLE 1 Computation of PRIDIT Scores Variable Label Proportion of B t1 Yes ("Yes") B t2 ( No") 1 44% % % % % % Claim Simple Example with Score and Class Ten Indicators (1= YES,2= NO ), Scores and Classes Score Class Claim PRIDIT Score for Treatment Variables PRIDIT Transformed Indicators, Scores and Classes Score Class Treatment was unusually prolonged (> 6 months) 7 24% Indep. Medical examiner questioned extent of treatment 8 11% Medical audit raised questions about charges 9 4%

13 Comparison: PRIDIT and Clustering PRIDIT gives a score, which may be very useful for claims sorting. Clustering assigns claims to classes. They are either in or out of the assigned class. Clustering creates border problems Clustering ignores information about the order of values for categorical variables Clustering can accommodate both categorical and continuous variables Clustering Comparison Unordered categorical variables with many values (i.e., injury type): Kahonen clustering is a procedure for measuring similarity/dissimilarity for these variables and in clustering If the values for the variables contain no meaningful order, PRIDIT will not help in creating variables to use in Principal Components Analysis. Application Result There appears to be a strong relationship between PRIDIT score and expert suspicion that claim is fraudulent or abusive The clusters resulting from the cluster procedure also appeared to be effective in separating legitimate from fraudulent or abusive claims The relative importance of variables may be very different from supervised regressions.

14 REFERENCES Derrig, R. A., (2002), Insurance Fraud, Journal of Risk and Insurance,, 69:3, Derrig, R. A., L. K. Krauss (1994), First Steps to Fight Workers Compensation Fraud, Journal of Insurance Regulation,, 12:3, Francis, L. and Derrig, R. A., (2008) Distinguishing the Forest from the TREES, A Comparison of Tree-Based Data Mining Methods, VARIANCE,, 2:2, Johnston, D.J., Derrig, R A., Sprinkel, E.A.,(2006) Auto Insurance Fraud: Measurements and Efforts to Combat It, Risk Management and Insurance Review, 9:2, Weisberg H.I., Derrig, R.A. (1992), Massachusetts Automobile Bodily Injury Tort Reform, Journal of Insurance Regulation, 10:2, REFERENCES Ai, J., Brockett, Patrick L., and Golden, Linda L. (2009) Assessing Consumer Fraud Risk in Insurance Claims with Discrete and Continuous Data, North American Actuarial Journal 13: Brockett, Patrick L., Derrig, Richard A., Golden, Linda L., Levine, Albert and Alpert, Mark, (2002), Fraud Classification Using Principal Component Analysis of RIDITs, Journal of Risk and Insurance,, 69:3, Brockett, Patrick L., Xiaohua, Xia and Derrig, Richard A., (1998), Using Kohonen Self-Organizing Feature Map to Uncover Automobile Bodily Injury Claims Fraud, Journal of Risk and Insurance,, 65: Bross, Irwin D.J., (1958), How To Use RIDIT Analysis, Biometrics, 4: Lieberthal, Robert D., (2010), Hospital Quality: A PRIDIT Approach, Health Services Research, Health Research and Educational Trust, PP 1-20 FUZZY REFERENCES Insurance Cummins, J.D. and Derrig, R.A. (1993), Fuzzy Trends in Property-Liability Insurance Claim Costs, Journal of Risk and Insurance,, September 1993, 60, pp Derrig, R.A. and Ostaszewski, K.M.(1995), Fuzzy Techniques of Pattern Recognition in Risk and Claim Classification, Journal of Risk and Insurance,, September, 62: Ostaszewski, K.M. (1993), An Investigation into Possible Applications of Fuzzy Sets Methods in Actuarial Science, Society of Actuaries DeWit, G.W. (1982), Underwriting and uncertainty, Insurance: Mathematics and Economics 1, pp Lemaire, Jean (1990), Fuzzy Insurance, Astin Bulletin 20(1), pp Young, V. (1994), Application of fuzzy sets to group health underwriting, Transactions of the Society of Actuaries 45, pp Young, V. (1996), Rate changing: A fuzzy logic approach, Journal of Risk and Insurance, 63:3, pp