International Journal of Computer Applications (0975 8887) Volume 39 No. !

"e#ruar$ %0 %

A Reviewof FinancialAccountingFraudDetectionbasedon DataMiningTechniques

Anuj Sharma
Information Systems Area Indian Institute of Management,
Indore, India

Prabin Kumar Panigrahi

Information Systems Area Indian Institute of Management, Indore, India

ABSTRACT
With an upsurge in financial accounting fraud in the current economic scenario experienced, financial accounting fraud detection (FAFD) has become an emerging topic of great importance for academic, research and industries. The failure of internal auditing system of the organization in identifying the accounting frauds has lead to use of specialized procedures to detect financial accounting fraud, collecti e !no"n as forensic accounting. Data mining techni#ues are pro iding great aid in financial accounting fraud detection, since dealing "ith the large data olumes and complexities of financial data are big challenges for forensic accounting. This paper presents a comprehensi e re ie" of the literature on the application of data mining techni#ues for the detection of financial accounting fraud and proposes a frame"or! for data mining techni#ues based accounting fraud detection. The systematic and comprehensi e literature re ie" of the data mining techni#ues applicable to financial accounting fraud detection may pro ide a foundation to future research in this field. The findings of this re ie" sho" that data mining techni#ues li!e logistic models, neural net"or!s, $ayesian belief net"or!, and decision trees ha e been applied most extensi ely to pro ide primary solutions to the problems inherent in the detection and classification of fraudulent data.

financial statement, especially in estors and creditors *4,. The accounting fraud is executed by ma!ing falsified financial accounting statements "here the numbers are manipulated by o erstating assets, spurious entries related to sales and profit, misappropriation in taxes, or understating liabilities, debts, expenses or losses *+,. The accounting fraud is also defined by accounting professionals as /deliberate and improper manipulation of the recording of data in financial statements in order to achie e an operating profit of the company and appear better than it actually is *5,. 'conomically, financial fraud is becoming an increasingly serious problem and effecti e detecting accounting fraud has al"ays been an important but complex tas! for accounting professionals *6,. The internal auditing of financial matters in the companies has become an increasingly demanding acti ity and there are many e idence that 7boo! coo!ing8 accounting practices are "orld9"ide applied for doing financial frauds *:,. The detection of accounting fraud using traditional internal audit procedures is a difficult or sometimes an impossible tas! *;,. First, the auditors usually lac! the re#uired !no"ledge concerning the characteristics of accounting fraud. %econd, as the fraudulent manipulation of accounting data is so infre#uent, most of the auditors lac! the experience and expertise needed to detect and pre ent frauds. Finally, the other concern people of finance department li!e )hief Financial -fficer ()F-), financial managers and accountants are intentionally trying to decei e the internal or external auditors *+<,. While !no"ing the limitations of an audit, finance and accounting managers ha e concluded that traditional and standard auditing procedures are insufficient to detect frauds. These limitations of financial auditing suggest the need for additional automatic data analysis procedures and tools for the effecti e detection of falsified financial statements. Although the latest re ision auditing standards is enlarging auditors8 fraud detection responsibility, effecti e detecting accounting fraud has al"ays been a problem for accounting profession *4,. The failure of internal auditing system of the organization in identifying the accounting frauds has lead to use of specialized procedures to detect financial accounting fraud collecti e !no"n as forensic accounting. Forensic accounting plays a ital role in detecting these frauds "hich are difficult to find out in internal auditing by employing accounting, auditing, and in estigati e s!ills *++, +.,. Without the fail9safe financial fraud pre ention tools and procedures, the accounting fraud has become a business critical problem in current competiti e en ironment. Data &ining based financial fraud detection and fraud control, automates the "hole process and helps to reduce the manual "or! of screening and chec!ing arious statements. This area

General Terms
Fraud Detection, Financial Fraud, Financial %tatements.

Keywords
Financial Accounting Fraud, Fraud Detection, Data &ining.

1. INTRODUCTION
With an upsurge in financial accounting fraud in the current economic scenario experienced, financial accounting fraud detection (FAFD) ha e recei ed considerable attention from the in estors, academic researchers, media, the financial community and regulators. Due to some high profile financial frauds disco ered and reported at large companies li!e 'nron, (ucent, World)om and %atyam o er the last decade, the re#uirement of detecting, defining and reporting financial accounting fraud has increased *+,.
The -xford 'nglish Dictionary *., defines fraud as /"rongful or criminal deception intended to result in financial or personal gain. 0n academic literature fraud is defined as leading to the abuse of a profit organization1s system "ithout necessarily leading to direct legal conse#uences *2,. Although the literature is missing a uni ersally accepted definition of financial fraud, researcher has defined it as /a deliberate act that is contrary to la", rule, or policy "ith intent to obtain unauthorized financial benefit *3, and /intentional misstatements or omission of amount by decei ing users of

26

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % has become one of the !ey applications of data mining techni#ues in established industry or go ernment organizations *2,. Data mining is !no"n as gaining insights and identifying interesting patterns from the data stored in large databases in such a "ay that the patterns and insights are statistically reliable, pre iously un!no"n, and actionable *+2,. Data mining is also define as /a process that uses statistical, mathematical, artificial intelligence and machine learning techni#ues to extract and identify useful information and subse#uently gaining !no"ledge from a large database *+3, +4,. The blending point bet"een data mining and detecting accounting fraud is that, data mining as an ad anced analytical tool may assist the auditors in decision ma!ing and detecting fraud. The data mining techni#ues ha e the potential to sol e the contradiction bet"een effect and efficiency of fraud detection *4,. Data mining plays an important role in financial accounting fraud detection, as it is often applied to extract and disco er the hidden patterns in ery large collection of data *6,. An auditor can ne er become certain about the legitimacy of and intention behind a fraudulent transaction. )oncerning this reality, the most optimal and cost effecti e option is to find out enough e idences of fraud from the a ailable data using specialized mathematical and data processing algorithms *2,. There are many researches that describe the applicability of data mining algorithms in detecting accounting fraud. As the research about the applicability of data mining techni#ues for detection of financial accounting fraud is a promising field, many re ie" articles ha e published in conference proceedings or =ournal publications. %tatistical methods of detecting different types of frauds li!e credit card fraud, fraudulent money laundering, telecommunications fraud, etc., are re ie"ed in *+5,. Applications of data mining in stoc! mar!ets and ban!ruptcy predictions and related fraud detection ha e been sur eyed in *+6,. A sur ey of data mining9based fraud detection research is presented in *2,, including credit transaction fraud, telecoms subscription fraud, automobile insurance fraud, terrorist detection, financial crime detection, intrusion and spam detection. -thers researcher ha e re ie"ed insurance fraud *+:, and financial statement fraud *+,. This paper presents a comprehensi e re ie" of the research literature on the application of data mining techni#ues to detect financial accounting fraud. The paper proposes a frame"or! for data mining techni#ues based accounting fraud detection to help certified public accountants selecting suitable data and data mining technologies for detecting fraud. The systematic and comprehensi e literature re ie" of the data mining techni#ues applicable to financial accounting fraud detection may pro ide a foundation to future research in this field. The rest of the paper is organized as follo"s. %ection . describes classification of data mining techni#ues and applications for financial accounting fraud detection. %ection 2 pro ides distribution of the research literature as per the applications and techni#ues of data mining for the detection of financial accounting fraud. %ection 3 describes our frame"or! in more detail. %ection 4 outlines future direction of this "or!.

2. CLASSIFICATION OF DATA MINING T C!NI"U S FOR FRAUD D T CTION

0n this section, a graphical conceptual frame"or! is proposed for the a ailable literature on the applications of data mining techni#ues to financial accounting fraud detection. The classification frame"or!, "hich is sho"n in Fig. +, is based on a literature re ie" of existing !no"ledge on the nature of data mining research *+;,.<,, fraud detection research *+,2,+5,+6,+:,. A classification frame"or! for financial fraud is suggested in *6, based on the financial crime frame"or! of the >.%. Federal $ureau of 0n estigation *.+,, "hich is one of the established frame"or!s for financial fraud Detection. Fig. + consists of t"o layers, the first comprising the six data mining application classes of classification, clustering, prediction, outlier detection, regression, and isualization *2,+5,+:,..,.2,, supported by a set of algorithmic approaches to extract the rele ant relationships in the data *+3,.

F#$. 1% T&e Con'e()*al Framewor+ ,or A((l#'a)#on o, Da)a M#n#n$ )o FAFD A brief description of the conceptual frame"or! "ith references is pro ided and of the six data mining application classes (classification, clustering, outlier detection, prediction, regression and isualization), each component of is discussed in more detail in the follo"ing sections.

2.1 Class#,#'a)#on A((l#'a)#ons

o,

Da)a

M#n#n$

'ach of the six data mining application classes is supported by a set of algorithmic approaches to extract the rele ant relationships in the data. These approaches can handle different classes of problems. The classes are presented belo".
Class#,#'a)#on 9 )lassification builds up (from the training set) and utilizes a model (on the target set) to predict the categorical labels of un!no"n ob=ects to distinguish bet"een ob=ects of different classes. These categorical labels are predefined, discrete and unordered *.3,. The research literature describes that classification or prediction is the process of identifying a set of common features (patterns), and proposing models that describe and distinguish data classes or concepts *+6,. )ommon classification techni#ues include neural net"or!s, the ?a@ e $ayes techni#ue, decision trees and support ector machines. %uch classification tas!s are used in the detection of credit card, healthcare and automobile insurance, and corporate fraud, among other types of fraud,

2:

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % and classification is one of the most common learning models in the application of data mining in fraud detection. Cl*s)er#n$ 9 )lustering is used to partition ob=ects into pre iously un!no"n conceptually meaningful groups (i.e. clusters), "ith the ob=ects in a cluster being similar to one another but ery dissimilar to the ob=ects in other clusters. )lustering is also !no"n as data segmentation or partitioning and is regarded as a ariant of unsuper ised classification *.3,. )luster analysis decomposes or partitions a data set (single or multi ariate) into dissimilar groups so that the data points in one group are similar to each other and are as different as possible from the data points in other groups *+,. 0t is suggested that data ob=ects in each cluster should ha e high intra9cluster similarity "ithin the same cluster but should ha e lo" inter9cluster similarity to those in other clusters *+6,. The most common clustering techni#ues are the A9nearest neighbour, the ?a@ e $ayes techni#ue and self9organizing maps. -red#')#on 9 Brediction estimates numeric and ordered future alues based on the patterns of a data set *+;,. 0t is noted that, for prediction, the attribute, for "hich the alue being predicted is continuous9 alued (ordered) rather than categorical (discrete9 alued and unordered). This attribute is referred as the predicted attribute *.3,. ?eural net"or!s and logistic model prediction are the most commonly used prediction techni#ues. O*)l#er De)e')#on 9 -utlier detection is employed to measure the /distance bet"een data ob=ects to detect those ob=ects that are grossly different from or inconsistent "ith the remaining data set *.3,. Data that appear to ha e different characteristics than the rest of the population are called outliers *.5,. The problem of outlierCanomaly detection is one of the most fundamental issues in data mining. A commonly used techni#ue in outlier detection is the discounting learning algorithm *.6,. Re$ress#on 9 Degression is a statistical methodology used to re eal the relationship bet"een one or more independent ariables and a dependent ariable (that is continuous9 alued) *.3,. &any empirical studies ha e used logistic regression as a benchmar! *.:,. The regression techni#ue is typically underta!en using such mathematical methods as logistic regression and linear regression, and it is used in the detection of credit card, crop and automobile insurance, and corporate fraud. .#s*al#/a)#on 9 Eisualization refers to the easily understandable presentation of data and to methodology that con erts complicated data characteristics into clear patterns to allo" users to ie" the complex patterns or relationships unco ered in the data mining process *+3,. The researchers ha e exploited the pattern detection capabilities of the human isual system by building a suite of tools and applications that flexibly encode data using colour, position, size and other isual characteristics. Eisualization is best used to deli er complex patterns through the clear presentation of data or functions *.;,. techni#ues are logistic models, neural net"or!s, the $ayesian belief net"or!, and decision trees, all of "hich fall into the /classification category. These four techni#ues are discussed in more detail in the follo"ing paragraphs. Re$ress#on Models 9 The regression based models are mostly used in financial accounting fraud detection. The ma=ority of them are based on logistic regression, step"ise9logistic regression, multi criteria decision ma!ing method and exponential generalized beta t"o ('F$.) *6,. (ogistic model is a generalized linear model that is used for binomial regression in "hich the predictor ariables can be either numerical or categorical *2<, 2+,. 0t is principally used to sol e problems caused by insurance and corporate fraud. %ome of the research has suggested logistic regression based model to predict the presence of financial statement fraud *2<, 22,. %tatistical method of logistic regression can detect falsified financial statements efficiently *2<,. %ome researchers ha e also de eloped generalized #ualitati e response model based on Brobit and (ogit techni#ues to predict financial statement fraud. That model "as based on a dataset collected by an international public accounting company and needs testing for generalization *23,. )ascaded (ogit model has also proposed to in estigate the relationship bet"een insider trading and possibility of fraud. The study in *24, found that, "hen the fraud is being executed, insiders, i.e. top executi es and managers, reduce their stoc! holdings through high stoc! selling acti ity. The other methods li!e statistical regression analysis are also useful to test if the existence of an independent audit committee mitigates or reduces the li!elihood of fraud. (iterature also describes that organizations "ith audit committees, formed by independent managers, meeting no more than t"ice per year, are less li!ely to be sanctioned for fraudulent financial reporting *25,. 0n .<<<, $ell and )arcello *26, proposed a logistic regression model for estimating the li!elihood of fraudulent financial reporting for an audit client. The model "as conditioned on the presence or absence of se eral fraud9ris! factors. The fraud ris! factors identified in the final model included "ea! internal control system, rapid company gro"th, inade#uate or inconsistent relati e profitability, management that =ust "ant to achie e earnings pro=ections anyho" "hile lying to the auditors or is o erly e asi e, company o"nership status (public s. pri ate), and interaction term bet"een a "ea! control en ironment and an aggressi e management attitude to"ards financial reporting. 0n .<<., %pathis et al. *2:, proposed that statistical techni#ues li!e logistic regression may be suitable to de elop a model to identify factors related to fraudulent financial statement. ?on9 parametric regression9based frame"or! "as used to run the falsified financial statement detection model. The proposed model "as compared "ith discriminant analysis and logit regression methods for benchmar!ing. The regression analysis using (ogit model can be used for empirical analysis of financial indexes "hich can significantly predict financial fraud *2;,. (ogistic analysis and clustering analysis =ointly can be used to establish a detecting model of fraud from four aspects of financial indexes, company go ernance, financial ris! and pressure and related trading. After cluster filtering significant ariables, prediction model can be established "ith methods of %tandardization, non9 %tandardization $ayes and (ogistic *3<,. The logistic regression based accounting fraud detecting models are common in literature since the model based on

2.2 Class#,#'a)#on o, Da)a M#n#n$ Te'&n#0*es ,or F#nan'#al A''o*n)#n$ Fra*d De)e')#on
To determine the main algorithms used for financial accounting fraud detection, "e present a De ie" of data mining techni#ues identified in literature applied to the detection of financial fraud. The most fre#uently used

2;

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % logistic regression can reach up to ;4.+G of detecting accuracy "ith significant expectation effect *3+,. Ne*ral Ne)wor+s H The neural net"or!s are non9linear statistical data modeling tools that are inspired by the functionality of the human brain using a set of interconnected nodes *2+, 2.,. ?eural net"or!s are "idely applied in classification and clustering, and its ad antages are as follo"s. First, it is adapti eI second, it can generate robust modelsI and third, the classification process can be modified if ne" training "eights are set. ?eural net"or!s are chiefly applied to credit card, automobile insurance and corporate fraud. (iterature describes that neural net"or!s can be used as a financial fraud detection tool. The neural net"or! fraud classification model employing endogenous financial data created from the learned beha ior pattern can be applied to a test sample *3.,. The neural net"or!s can be used to predict the occurrence of corporate fraud at the management le el *32,. Desearchers ha e explored the effecti eness of neural net"or!s, decision trees and $ayesian belief net"or!s in detecting fraudulent financial statements (FF%) and to identify factors associated "ith FF% *:,. The study in *+<, re ealed that input ector consisted of financial ratios and #ualitati e ariables, "as more effecti e "hen fraud detection model "as de eloped using neural net"or!. The model "as also compared "ith standard statistical methods li!e linear and #uadratic discriminant analysis, as "ell as logistic regression methods *+<,. The generalized adapti e neural net"or! architectures and the adapti e logic net"or! are "ell recei ed for fraud detection. The hybrid techni#ues li!e fuzzy rule integrated "ith a neural net"or! (neuro9fuzzy systems) are also proposed. The literature describes that the integrated fuzzy neural net"or! outperformed traditional statistical models and neural net"or!s models reported in prior studies *33,. Bayes#an Bel#e, Ne)wor+ 9 The $ayesian belief net"or! ($$?) represents a set of random ariables and their conditional independencies using a directed acyclic graph (DAF), in "hich nodes represent random ariables and missing edges encode conditional independencies bet"een the ariables *:,. The $ayesian belief net"or! is used in de eloping models for credit card, automobile insurance, and corporate fraud detection. The research in *:, described that $ayesian belief net"or! model correctly classified ;<.2G of the alidation sample for fraud detection. $ayesian belief net"or! outperformed neural net"or! and decision tree methods and achie ed outstanding classification accuracy *:,. De'#s#on Trees H A decision tree (DT) is a tree structured decision support tool, "here each node represents a test on an attribute and each branch represents possible conse#uences. 0n this "ay, the predicti e model attempts to di ide obser ations into mutually exclusi e subgroups and is used for data mining and machine learning tas!s *:,. Decision trees are predicti e decision support tools that create mapping from obser ations to possible conse#uences *.3,. These trees can be planted ia machine9learning9based algorithms such as the 0D2, )ADT and )3.4. Bredictions are represented by lea es and the con=unctions of features by branches. Decision trees are commonly used in credit card, automobile insurance, and corporate fraud. To identify and predict the impact of fraudulent financial statements, classification and regression trees ()ADT) algorithm is introduced in *45,. Na12e Bayes 9 ?a@ e $ayes is used as simple probabilistic classifier based on $ayes conditional probability rule. ?a@ e $ayes follo"s strong (nai e) statistical independence assumptions for the predictor ariables. 0t is an effecti e classification tool that is easy to interpret and particularly suited "hen the dimensionality of the inputs is high *.3,. 0n a study *2,, ?a@ e $ayes classifier outperformed the con entional classifier. The efficiency of predicting financial fraud data "as higher "ith no false positi es "ith relati ely lo" false negati es. ?a@ e $ayes methods are "idely used in ban!ing and financial fraud detection and claim fraud detection. To apply Ada $oosted ?a@ e $ayes scoring to insurance claims fraud, a case study is gi en to diagnosis claim fraud *34,. Neares) Ne#$&3o*r Me)&od 9 ?earest neighbour method is a similarity based classification approach. $ased on a combination of the classes of the most similar ! record(s), e ery record is classified. %ometimes this method is also !no"n as the !9nearest neighbour techni#ue *.3,. A9nearest neighbour method is used in automobile insurance claims fraud detection *35, and for identifying defaults of credit card clients *2+,. F*//y lo$#' and Gene)#' Al$or#)&m H Fenetic algorithms are used in classifier systems to represent and modeling the auditor decision beha ior in a fraud setting *36,. Fenetic algorithm along "ith binary support ector system ($%E%) "hich is based on the support ectors in support ector machines (%E&) are used to sol e problems of credit card fraud that had not been "ell identified *3:,. Fuzzy (ogic is a mathematical techni#ue that classifies sub=ecti e reasoning and assigns data to a particular group, or cluster, based on the degree of possibility the data has of being in that group. The expert fuzzy classification techni#ues enable one to perform approximate reasoning that can impro e performance in three "ays. First, performance is impro ed through efficient numerical representation of ague terms, because the fuzzy technology can numerically sho" representation of a data item in a particular category. The second "ay performance is enhanced is through increased range of operation in ill9defined en ironments, "hich is the "ay that fuzzy methodology can sho" partial membership of data elements in one or more categories that may not be clearly defined in traditional analysis. Finally, performance is increased because the fuzzy technology has decreased sensiti ity to JnoisyJ data, or outliers. A multile el fuzzy rule9 based system is proposed in *3;, to ran! state financial management. The authors used fuzzy set theory to represent imprecision in e aluated information and =udgments. A fuzzy logic model has been implemented in *4<, for fraud detection in an 'xcel spreadsheet. $y using the fuzzy logic model to de elop clusters for different statements representing red flags in the detection of fraud, non9financial data "as included "ith financial statement ariables for the analysis. The model consist of different financial ariables li!e le erage, profitability, li#uidity, cash flo" and a ariable designed to represent a company8s ris! of fraud. Fuzzy logic efficiently modeled the ariable, "hich "as de eloped to #uantify fraud ris! factors. The model predicted frauds "ith :5.6G accuracy *4<,. The same model "as adapted in *4+, to de elop a model for detection of financial statement fraud. The proposed model used a combination of different financial statement data. The data included in the model consisted of easily a ailable non9financial information, and a specialized ariable constructed by #uantifying the company8s financial statements issues using fuzzy logic. The study re ealed that

3<

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % most successful predictor ariables "ere financial ariables indicating cash flo", cash li#uidity, the ratio of sales to assets, and a ariable indicating company size. The model predicted frauds "ith 65.6 G accuracy *4+,. Fuzzy logic based expert system has been de eloped to identify and e aluate "hether elements of fraud are in ol ed in insurance claims settlements. The fuzzy logic based expert system "as de eloped for auditors to identify fraud in settled claimed insurance. The system "as able to cut costs by detecting fraudulent filings *4.,. The other fuzzy logic based fraud detection systems are gi en in *46, 4:,. Fenetic programming "ith fuzzy logic production rules is used to classifying data. The study in *42, has proposed and tested a system to detect frauds on real home insurance claims and credit card transaction data. The study on genetic programming for fraud detection lac!s benchmar!ing "ith the existing methods and techni#ues. A genetic algorithm based approach to detect financial statement fraud in presented in *43,. 0t "as found that exceptional anomaly scores are aluable metrics for characterizing corporate financial beha ior and that analyzing these scores o er time represents an effecti e "ay of detecting potentially fraudulent beha ior. Another study on fraud detection applied many different classification techni#ues in a super ised t"o9class setting and a semi9super ised one9class setting in order to compare the performances of these techni#ues and settings *44,. 4(er) Sys)ems H Desearchers in the field of 'xpert systems ha e examined the role of 'xpert %ystems in increasing the detecting ability of auditors and statement users. $y using expert system, they could ha e better detecting abilities to accounting fraud ris! under different context and le el and enable auditors gi e much reliable auditing suggestions through rational auditing procedure. The research has confirmed that the use of an expert system enhanced the auditors8 performance. With assistance from expert system, the auditors discriminated better, among situations "ith different le els of management fraud9ris!. 'xpert %ystem aided in decision ma!ing regarding appropriate audit actions *4;,. The financial accounting fraud detection research is classified as per data mining application and data mining techni#ues. %ome researchers ha e tried to apply a combination of many data mining techni#ues li!e decision trees, neural net"or!s, $ayesian belief net"or!, A9nearest neighbour as in *5<,. The main ob=ecti e is to apply a hybrid decision support system using stac!ing ariant methodology to detect fraudulent financial statements. %ome research is targeted to indentify financial frauds using simple logistic regression models in specific country li!e in ?e" Kealand *5+, and )hina *5.,. %ome of the recent study in the financial fraud detection ma!e used of t"o or more data mining applications in a hybrid manner or =ust attempt to compare their effecti eness *6<96.,. expert system and genetic algorithm for financial accounting fraud detection.

7. DATA MINING BAS D FRAM 6ORK FOR FRAUD D T CTION

The research related "ith application of data mining algorithms and techni#ues for financial accounting fraud detection is a "ell studied area. The implementation of these techni#ues follo"s the same information flo" of data mining processes in general. The process starts "ith feature selection then proceeds "ith representation, data collection and management, pre9processing, data mining, post9processing, and in the end performance e aluation. This paper has proposed an expanded generic data mining frame"or!. This frame"or! considers specific characteristics of fraud detection techni#ues for financial accounting fraud (see Fig. .).

8. CONCLUSION AND FUTUR R S ARC!

This paper re ie"ed the literature describing use of data mining algorithms including statistical test, regression analysis, ?eural ?et"or!, decision tree, $ayesian net"or! etc for financial accounting fraud detection. Degression Analysis is "idely used for fraud detection since it has great explanation ability. Different regression model used by researchers are (ogit, %tep9"ise (ogistic, >TAD0% and 'F$. etc. ?eural ?et"or!s are important tool for data mining. The researchers ha e not made any comparison so far, related "ith detecting effect and accuracy of ?eural ?et"or! compared to regression model. The ad antages of ?eural ?et"or! are that there are no strict re#uests for data and it has a strong generalization and ad=ustment. After correct allocation and proper training, ?eural ?et"or! may perform great classification comparing "ith regression model. $ut due to special inner hidden structure, it is impossible for researchers to trac! the formation process of output conclusion. There are other issues also related "ith ?eural ?et"or! li!e no clear explanation on connecting "eight, complex accuracy and statistical reliability chec!ing procedure, and lac! of explanation. This paper suggests that using only financial statements data may not be sufficient for detections of fraud. The importance of data mining techni#ues in the detection of financial fraud has been recognized. The future "or! may be proposing a comprehensi e classification frame"or! or a systematic re ie" of data mining application in financial accounting fraud detection. 0n this study, "e conduct an extensi e re ie" of academic articles and pro ide a comprehensi e bibliography and classification frame"or! for the applications of data mining to Financial Accounting Fraud Detection. The intention is to inform both academics and practitioners of the areas in "hich specific data mining techni#ues can be applied to financial accounting fraud detection, and to report and compile a systematic re ie" of the burgeoning literature on financial accounting fraud detection. Although our study cannot claim to be exhausti e, "e belie e that it "ill pro e a useful resource for anyone interested in financial accounting fraud detection research, and "ill help simulate further interest in the field.

5. T! R S ARC! UND R -RO-OS D CLASSIFICATION FRAM 6ORK

The Table +, ., 2 and 3 sho" the distribution of the research papers as per the order of the year of publication. The Table + presents the research "or! on neural net"or! for financial accounting fraud detection. The literature re ie" is presented in the order of publication. The Table . presents the research "or! on regression models for financial accounting fraud detection. The Table 2 sho"s the research "or! on fuzzy logic and Table 3 presents research "or! on

3+

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % Ta3le 1% Resear'& on Ne*ral Ne)wor+ ,or F#nan'#al A''o*n)#n$ Fra*d De)e')#on S. No. + . 2 3 4 5 6 A*)&or Fanning, )ogger and %ri asta a (+;;4) Freen and )hoi (+;;6) Fanning and )ogger (+;;:) )erullo and )erullo (+;;;) Aos!i aara (.<<<) Feroz et al. (.<<<) (in, L"ang and $ec!er (.<<2) Aotsiantis et al. (.<<5) Air!os et al. (.<<6) Fen9&ay (iou (.<<:) & Arambia9 Aapardis et al. (.<+<) Da isan!ar et al. (.<++) Berols (.<++) Khou and Aapoor (.<++) Da)a M#n#n$ Te'&n#0*es ?eural ?et"or!s ?eural ?et"or!s ?eural ?et"or!s ?eural ?et"or!s ?eural ?et"or!s ?eural ?et"or!s Fuzzy ?eural ?et"or!, (ogistic &odel Decision Trees, ?eural ?et"or!s, $ayesian $elief ?et"or!, A9 ?earest ?eighbour ?eural ?et"or!s, Decision Trees, $ayesian $elief ?et"or! ?eural ?et"or!s Ma#n O39e')#2e To use neural net"or!s to de elop a model for detecting managerial fraud To de elop a neural net"or! fraud classification model employing endogenous financial data in corporate fraud To use neural net"or!s to de elop a model for detecting managerial fraud To use neural net"or!s to predict the occurrence of corporate fraud at the management le el To in estigate the impact of arious pre9processing models on the forecast capability of neural net"or! for auditing financial accounts To predict the possible fraudsters and accounting manipulations To e aluate the utility of an integrated fuzzy neural net"or! model for corporate fraud detection To apply a hybrid decision support system using stac!ing ariant methodology to detect fraudulent financial statements To explore the effecti eness of neural net"or!s, decision trees and $ayesian belief net"or!s in detecting fraudulent financial statements (FF%) and to identify factors associated "ith FF% To build detectionCprediction models for detecting fraudulent financial reporting To test the use of artificial neural net"or!s as a tool in fraud detection To identify companies that resort to financial statement fraud To compares the performance of popular statistical and machine learning models in detecting financial statement fraud To detect financial statement fraud "ith exploring a self9adapti e frame"or! (based on a response surface model) "ith domain !no"ledge Re,eren'e *;, *3., *+<, *32, *54, *55, *33,

*5<,

*:,

+< ++ +. +2 +3

*5:, *5;, *6<, *6+, *6.,

?eural ?et"or!s ?eural ?et"or!, %upport Eector &achines ?eural ?et"or!s, %upport Eector &achines ?eural ?et"or!s, $ayesian ?et"or!s

Ta3le 2% Resear'& on Re$ress#on Models ,or F#nan'#al A''o*n)#n$ Fra*d De)e')#on S. No. + . 2 4 3 5 A*)&or Bersons (+;;4) $easley (+;;5) Lansen et al. (+;;5) %ummers and %"eeney (+;;:) Abbot, Bar! and Bar!er (.<<<) $ell and )arcello (.<<<) Da)a M#n#n$ Te'&n#0*es (ogistic &odel (ogit Degression Analysis Brobit And (ogit Techni#ues (ogit Degression Analysis %tatistical Degression Analysis (ogistic &odel Ma#n O39e')#2e To detect financial reporting frauds To predict the presence of financial statement fraud To use Brobit and (ogit techni#ues to predict fraud To in estigate the relationship bet"een insider trading and fraud To examine if the existence of an independent audit committee mitigates the li!elihood of fraud To de elop a logistic regression model to estimate fraudulent financial reporting for an audit client Re,eren'e *.4, *22, *23, *24, *25, *26,

3.

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % To use logistic regression to examine published data and de elop a model to detect the factors associated "ith FF% To de elop a model to identify factors associated "ith fraudulent financial statement To explore the (ogit regression model to detect corporate fraud in ?e" Kealand To analyze financial indexes "hich can predict financial fraud To establish a detecting model of fraud "hich can be used for empirical analysis of financial indexes To de elop accounting fraud detecting model To introduce classification and regression trees to identify and predict the impact of fraudulent financial statements To employ a logistic regression model to test the effects of managerial compensation and mar!et competition on financial fraud among listed companies in )hina To build detectionCprediction models for detecting fraudulent financial reporting To identify companies that resort to financial statement fraud To compares the performance of popular statistical and machine learning models in detecting financial statement fraud To detect financial statement fraud

6 : ; +< ++ +. +2

%pathis (.<<.) %pathis, Doumpos, and Kopounidis (.<<.) -"usu9Ansah et al. (.<<.) Muemin Luang (.<<5) Laisong Den (.<<5) Fuoxin et al. (.<<6) $ai, Nen and Nang (.<<:) Nuan et al. (.<<:)

(ogistic Degression (ogistic Degression (ogistic Degression &odels Degression Analysis >sing (ogit &odel (ogistic Analysis And )lustering Analysis (ogistic Degression )lassification And Degression Trees ()ADT) (ogistic Degression &odels (ogistic Degression, )lassification Trees (ogistic Degression (ogistic Degression, )3.4 Degression, Decision Tree

*2<, *2:, *5+, *2;, *3<, *3+, *45,

+3

*5.,

+4 +5 +6 +:

Fen9&ay (iou (.<<:) Da isan!ar et al. (.<++) Berols (.<++) Khou and Aapoor (.<++)

*5:, *6<, *6+, *6.,

Ta3le 5% Resear'& on F*//y Lo$#' ,or F#nan'#al A''o*n)#n$ Fra*d De)e')#on S. No. + . 2 3 4 5 6 A*)&or Deshmu!h, Domine and %iegel (+;;6) Deshmu!h and Talluru (+;;:) Ammar, Wright and %elden (.<<<) (enard and Alam (.<<3) Batha!, Eidyarthi and %ummers (.<<4) )hai, Loogs and Eerschueren (.<<5) (enard, Wat!ins and Alam (.<<6) Da)a M#n#n$ Te'&n#0*es Fuzzy (ogic Dule9$ased Fuzzy Deasoning %ystem Fuzzy (ogic Fuzzy (ogic and 'xpert Deasoning Fuzzy (ogic and 'xpert %ystem Fuzzy (ogic Fuzzy (ogic Ma#n O39e')#2e To pro ide a fuzzy sets model to assess the ris! of managerial fraud To build a rule9based fuzzy reasoning system to assess the ris! of managerial fraud To use fuzzy set theory to represent imprecision in e aluated information and =udgments To de elop fuzzy logic model to de elop clusters for different statements representing red flags in the detection of fraud To identify fraud in settled claims To con ert binary classification rules learned from a genetic Algorithm to a fuzzy score for financial data fraud rule matching To detect financial statement fraud using fuzzy logic Re,eren'e *4:, *46, *3;, *4<, *4., *52, *4+,

Ta3le 7% Resear'& on 4(er) Sys)em and Gene)#' Al$or#)&m ,or F#nan'#al A''o*n)#n$ Fra*d De)e')#on S. No. + . A*)&or Bacheco et al. (+;;5) 'ining, Oones and (oebbec!e (+;;6) Da)a M#n#n$ Te'&n#0*es Lybrid intelligent system "ith ?? and fuzzy expert system 'xpert %ystem Ma#n O39e')#2e To diagnose financial problems in companies To build an expert system applying the (ogit statistical model to enhance user engagement and Re,eren'e *56, *4;,

4 3

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % increase reliance on the aid Welch, Dee es and Welch (+;;:) Aiehl, Loogs and (a)omb (.<<4) Loogs et al. (.<<6) Ouszcza! et al. (.<<:) ' olutionary algorithms (genetic algorithms) Fenetic Algorithm Fenetic Algorithm %uper ised and %emi9 %uper ised )lassification To use genetic algorithms to aid the decisions of Defense )ontractor Audit Agency (D)AA) auditors "hen they are estimating the li!elihood of contracts fraud To automatically detect financial statement fraud To detect financial statement fraud based on anomaly scores as a metrics for characterizing corporate financial beha ior To detect financial statement fraud

2 3 4 5

*36, *53, *43, *44,

Fea)*re Sele')#on and Re(resen)a)#on Da)a So*r'es Financial %tatements Audit Deports 0nternal )ontrol Account (edgers

Da)a Clean#n$ and In)e$ra)#on Tra#n#n$: Tes) Da)a Se) Fraud Firm ?on9Fraud Firms DM Te'&n#0*es Da)a M#n#n$ ;DM< Degression ?eural ?et"or!s $ayesian $elief ?et"or! ?a@ e $ayes ?earest ?eighbour Fuzzy logic Decision Tree Fenetic Algorithm 'xpert %ystem

Da)a Sele')#on and Trans,orma)#on DM A((l#'a)#on )lassification )lustering Brediction -utlier detection Degression Eisualization

-a))ern 2al*a)#on

-os) -ro'ess#n$ o, -a))erns and Trends

-er,orman'e 2al*a)#on

Me)r#'s 'rror Date

F#$. 2% T&e Da)a M#n#n$ Based Framewor+ ,or F#nan'#al A''o*n)#n$ Fra*d De)e')#on

33

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % The data mining techni#ues of outlier detection and isualization ha e seen only limited use. The lac! of research on the application of outlier detection techni#ues to financial accounting fraud detection may be due to the difficulty of detecting outliers. 0ndeed, research has point out that outlier detection is a ery complex tas! a!in to finding a needle in a haystac!. Distinct from other data mining techni#ues, outlier detection techni#ues are dedicated to finding rare patterns associated "ith ery fe" data ob=ects. 0n the field of financial accounting fraud detection, outlier detection is highly suitable for distinguishing fraudulent data from authentic data, and thus deser es more in estigation. %imilarly, isualization techni#ues ha e a strong ability to recognize and present data anomalies, "hich could ma!e the identification and #uantification of fraud schemes much easier. The paper suggests that one of the reasons for the limited number of rele ant =ournal articles published for financial accounting fraud detection is the difficulty of obtaining sufficient research data. Fanning and )ogger *;, highlight the challenge of obtaining fraudulent financial statements, and note that this creates enormous obstacles in financial accounting fraud detection research. The most urgent challenge facing financial accounting fraud detection is to bridge the gap bet"een practitioners and researchers. The existing financial accounting fraud detection research concentrates on particular types of data mining techni#ues or models, but future research should direct its attention to"ard finding more practical principles and solutions for practitioners to help them to design, de elop, and implement data mining and business intelligence systems that can be applied to financial accounting fraud detection. We predict that increasing amounts of pri acy9preser ing financial data "ill be publicly a ailable in the near future due to increased collaboration bet"een practitioners and researchers, and that this should lead to more in estigations of data mining techni#ues that can be applied to pri acy9 preser ing data. A further problem faced by financial accounting fraud detection is that of cost sensiti ity. The cost of misclassification (false positi e and false negati e errors) differs, "ith a false negati e error (misclassifying a fraudulent acti ity as a normal acti ity) usually being more costly than a false positi e error (misclassifying a normal acti ity as a fraudulent acti ity) *2,. Fe" studies ha e explicitly included cost in their financial accounting fraud detection modeling *.:,, but future research on the application of data mining techni#ues to financial accounting fraud detection problems should ta!e into account cost sensiti ity considerations. This study has t"o ma=or limitations. First, our re ie" applied se eral !ey"ords to search only some online databases for articles published bet"een +;;. and .<++. A future re ie" could be expanded in scope. %econd, "e considered only articles for financial accounting fraud. Future research could be expanded to include rele ant data mining application for detection of frauds in other area li!e health, communication, insurance, ban!ing etc. *., -xford )oncise 'nglish Dictionary, ++th 'dition, -xford >ni ersity Bress, .<<;. *2, Bhua, )., (ee, E., %mith, A. P Fayler, D. (.<<4). A comprehensi e sur ey of data mining9based fraud detection research, Artificial 0ntelligence De ie" (.<<4) +H+3. *3, Wang, O., (iao, N., Tsai, T. P Lung, F. (.<<5). Technology9based financial frauds in Tai"anQ issue and approaches, 0''' )onference onQ %ystems, &an and )yberspace -ct (.<<5) ++.<H++.3. *4, Wang, %. (.<+<). A )omprehensi e %ur ey of Data &ining9$ased Accounting9Fraud Detection Desearch. 0nternational )onference on 0ntelligent )omputation Technology and Automation, ol. +, pp.4<942, .<+<. *5, Accounting Fraud Definition and 'xamples retrie ed from httpQCC""".accountingelite.comCaccounting9 tipsCaccounting9fraud9definition9and9examples9free9 accounting9fraud9articleC *6, ?gai, '.W.T., Lu, N., Wong, N. L., )hen, N., P %un, M. (.<+<). The application of data mining techni#ues in financial fraud detectionQ A classification frame"or! and an academic re ie" of literature, Decision %upport %ystem (.<+<), doiQ+<.+<+5C=.dss..<+<.<:.<<5. *:, Air!os, '., %pathis, )., P &anolopoulos, N. (.<<6). Data mining techni#ues for the detection of fraudulent financial statements, 'xpert %ystems "ith Applications 2. (3) (.<<6) ;;4H+<<2. *;, Fanning, A., )ogger, A., P %ri asta a, D. (+;;4). Detection of management fraudQ a neural net"or! approach. 0nternational Oournal of 0ntelligent %ystems in Accounting, Finance P &anagement, ol. 3, no. ., pp. ++2H .5, Oune +;;4. *+<, Fanning, A., P )ogger, A. (+;;:). ?eural net"or! detection of management fraud using published financial data. 0nternational Oournal of 0ntelligent %ystems in Accounting, Finance P &anagement, ol. 6, no. +, pp. .+9 .3, +;;:. *++, %il erstone, Lo"ard, P %heetz, &. (.<<3). Forensic Accounting and Fraud 0n estigation for ?on9'xperts. Lobo!en, Oohn Wiley P %ons, .<<3. *+., $ologna, Oac! P (ind#uist, D. O. (+;:6). Fraud Auditing and Forensic Accounting. ?e" Nor!Q Oohn Wiley P %ons, +;:6. *+2, 'l!an, ). (.<<+). &agical Thin!ing in Data &iningQ (essons from )-0( )hallenge .<<<. Broc. of %0FADD<+, 3.5932+. *+3, Turban, '., Aronson, O.'., (iang, T.B., P %harda, D. (.<<6). Decision %upport and $usiness 0ntelligence %ystems, 'ighth edition, Bearson 'ducation, .<<6. *+4, Fra"ley, W. O., Biatets!y9%hapiro, F., P &atheus, ).O. (+;;.). Ano"ledge disco ery in databasesQ An o er ie", A0 &agazine +2 (2) (+;;.) 46H6<. *+5, $olton, D. O., P Land, D.O. (.<<.). %tatistical fraud detectionQ a re ie", %tatistical %cience +6 (2) (.<<.) .24H.44. *+6, Khang, D., P Khou, (. (.<<3). Disco ering golden nuggetsQ data mining in financial application, 0''' Transactions on %ystems, &an and )ybernetics 23 (3) (.<<3) ?o . *+:, Derrig, D. A. (.<<.). 0nsurance fraud, The Oournal of Dis! and 0nsurance 5; (2) (.<<.) .6+H.:6.

=. R F R NC S
*+, Nue, M., Wu, N., Wang, N. (., P )hu, ). (.<<6). A re ie" of data mining9based financial fraud detection research, international conference on "ireless communications %ep, ?et"or!ing and &obile )omputing (.<<6) 44+;H44...

34

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % *+;, Ahmed, %.D. (.<<3). Applications of data mining in retail business, 0nternational )onference on 0nformation TechnologyQ )oding and )omputing . (.) (.<<3) 344H 34;. *.<, &itra, %., Bal, %.A., P &itra, B. (.<<.). Data mining in soft computing frame"or!Q a sur ey, 0''' Transactions on ?eural ?et"or!s +2 (+) (.<<.) 2H+3. *.+, F$0, Federal $ureau of 0n estigation, Financial )rimes Deport to the Bublic Fiscal Near, Department of Oustice, >nited %tates, .<<6, httpQCC""".fbi.go CpublicationsC financial CfcsR report.<<6CfinancialRcrimeR.<<6.htm. *.., Fa"cett, T., P Bro ost, F. (+;;6). Adapti e fraud detection, Data &ining and Ano"ledge Disco ery + (2) (+;;6) .;+H2+5. *.2, %Snchez, D., Eila, &.A., )erda, (., P %errano, O.&. (.<<;). Association rules applied to credit card fraud detection, 'xpert %ystems "ith Applications 25 (.) (.<<;) 252<H253<. *.3, Lan, O., P Aamber, &. (.<<5). Data &iningQ )oncepts and Techni#ues, %econd edition, &organ Aaufmann Bublishers, .<<5, pp. .:4H353. *.4, Bersons -.%. (+;;4). >sing Financial %tatement Data to 0dentify Factors Associated "ith Fraudulent Financial Deporting .Oournal of Applied $usiness Desearch,+;;4,++(2)Q2:935. *.5, Agyemang, &., $ar!er, A., P Alha==, D. (.<<5). A comprehensi e sur ey of numeric and symbolic outlier mining techni#ues, 0ntelligent Data Analysis +< (5) (.<<5) 4.+H42:. *.6, Namanishi, A., Ta!euchi, O., Williams, F., P &ilne, B. (.<<3). -n9line unsuper ised outlier detection using finite mixtures "ith discounting learning algorithms, Data &ining and Ano"ledge Disco ery : (2) (.<<3) .64H2<<. *.:, Eiaene, %., Derrig, D.A., $aesens, $., P Dedene, F. (.<<.). A comparison of state9of9the9art classification techni#ues for expert automobile insurance claim fraud detection, The Oournal of Dis! and 0nsurance 5; (2) (.<<.) 262H3.+. *.;, 'ic!, %.F. P Fyoc!, D.'. (+;;5). Eisualizing corporate data, ATPT Technical Oournal 64 (+) (+;;5) 63H:5. *2<, %pathis, ). T. (.<<.). Detecting false financial statements using published dataQ some e idence from Freece, &anagerial Auditing Oournal +6 (3) (.<<.) +6;H +;+. *2+, Neh, 0., P (ien, ). (.<<:). The comparisons of data mining techni#ues for the predicti e accuracy of probability of default of credit card clients, 'xpert %ystems "ith Applications 25 (.) (.<<:) .362H.3:<. *2., Fhosh, %., P Deilly, D. (. (+;;3). )redit card fraud detection "ith a neural9net"or!, .6th Annual La"aii 0nternational, )onference on %ystem %cience 2 (+;;3) 5.+H52<. *22, $easley, &. (+;;5). An empirical analysis of the relation bet"een board of director composition and financial statement fraud. The Accounting De ie", 6+(3), 332H 355. *23, Lansen, O. E., &cDonald, O. $., &essier, W. F., P $ell, T. $. (+;;5). A generalized #ualitati eTresponse model and the analysis of management fraud. &anagement %cience, 3.(6), +<..H+<2.. *24, %ummers, %. (., P %"eeney, O. T. (+;;:). Fraudulent misstated financial statements and insider tradingQ an empirical analysis. The Accounting De ie", 62(+), +2+H +35. *25, Abbot, O. (., Bar!, N., P Bar!er, %. (.<<<). The effects of audit committee acti ity and independence on corporate fraud. &anagerial Finance, .5(++), 44H56. *26, $ell, T., P )arcello, O. (.<<<). A decision aid for assessing the li!elihood of fraudulent financial reporting. AuditingQ A Oournal of Bractice P Theory, ;(+), +5;H +6:. *2:, %pathis, )., Doumpos, &., P Kopounidis, ). (.<<.). Detecting falsified financial statementsQ a comparati e study using multicriteria analysis and multi ariate statistical techni#ues. The 'uropean Accounting De ie", ++(2), 4<;H424. *2;, Luang, M. (.<<5). Desearch on Bublic )ompany Accounting Fraud and regulation9from perspecti e of protecting in estors *D,. MiamenQ Miamen >ni ersity, .<<5. *3<, Den, L. (.<<5). 0n estigation Desearch on Bublic company financial report fraud. *D, DalianQ Dongbei >ni ersity of Finance, .<<5. *3+, )hen, F., Khan=ia (., P Feng, L. (.<<6). 'mpirical %tudy on detecting financial statements Fraud9 based on empirical data of public companies. *O, Auditing %tudy, .<<6. *3., Freen, $. B., P )hoi, O. L. (+;;6). Assessing the ris! of management fraud through neural9net"or! technology. AuditingQ A Oournal of Bractice and Theory, +5(+), +3H .:. *32, )erullo, &. O., )erullo, E. (+;;;). >sing neural net"or!s to predict financial reporting fraud, )omputer Fraud P %ecurity &ayCOune (+;;;) +3H+6. *33, (in, O. W., L"ang, &. 0., P $ec!er, O. D. (.<<2). A Fuzzy ?eural ?et"or! for Assessing the Dis! of Fraudulent Financial Deporting. &anagerial Auditing Oournal, .<<2, +:(:)Q5469554. *34, Eiaene, %., Derrig, D. A., P Dedene F. (.<<3). A )ase %tudy of Applying $oosting ?ai e $ayes to )laim Fraud Diagnosis, 0''' Transactions on Ano"ledge and Data 'ngineering, .+5 n.4, p.5+.95.<, &ay .<<3. *35, Eiaene, %., Derrig, D. A., $aesens, P Dedene, F. (.<<.). A comparison of state9of9the9art classification techni#ues for expert automobile insurance claim fraud detection, The Oournal of Dis! and 0nsurance 5; (2) (.<<.) 262H 3.+. *36, Welch, O., Dee es, T. '., P Welch, %. T. (+;;:). >sing a genetic algorithm9based classifier system for modeling auditor decision beha ior in a fraud setting, 0nternational Oournal of 0ntelligent %ystems in Accounting, Finance P &anagement 6 (2) (+;;:) +62H+:5. *3:, )hen, D., )hen, T., P (in, ). (.<<5) A ne" binary support ector system for increasing detection rate of credit card fraud, 0nternational Oournal of Battern Decognition and Artificial 0ntelligence .< (.) (.<<5) ..6H.2;. *3;, Ammar, %al"a, Wright, D., P %elden, %. (.<<<). Dan!ing %tate Financial &anagementQ A &ultile el Fuzzy Dule9based %ystem. Decision %ciences. 2+Q 33;9 3:+. .<<<.

35

International Journal of Computer Applications (0975 8887) Volume 39 No. ! "e#ruar$ %0 % *4<, (enard, &. O., P Alam, B. (.<<3). The use of fuzzy logic and expert reasoning for !no"ledge management and disco ery of financial reporting fraud. 0n L. ?emati and ). $ar!o ('ds.), Lershey, BAQ 0dea Froup, 0nc. *4+, (enard, &. O., Wat!ins, A.(., and Alam, B. (.<<6). 'ffecti e use of integrated decision ma!ingQ An ad anced technology model for e aluating fraud in ser ice9based computer and technology firms. The Oournal of 'merging Technologies in Accounting 3(+)Q +.29+26. *4., Batha!, O., Eidyarthi, ?., P %ummers, %. (. (.<<4). A fuzzy9based algorithm for auditors to detect elements of fraud in settled insurance claims, &anagerial Auditing Oournal .< (5) (.<<4) 52.H533. *42, $entley, B. (.<<<). ' olutionary, my dear WatsonQ 0n estigating )ommittee9based ' olution of Fuzzy Dules for the Detection of %uspicious 0nsurance )laims. 0n the Broc. of F'))- .<<<. *43, Loogs, $., Aiehl, T., (acomb, )., P %entur!, D. (.<<6). A genetic algorithm approach to detecting temporal patterns indicati e of financial statement fraud, 0ntelligent %ystems in Accounting, Finance and &anagement, .<<6, ol. +4Q 3+945. *44, Ouszcza!, B., Adams, ?.&., Land, D.O., Whitro", )., P Weston, D.O. (.<<:). -ff9the9peg and bespo!e classifiers for fraud detection, )omputational %tatistics and Data Analysis, ol. 4. (;)Q 34.+9342.. *45, $ai, $., Nen, O. P Nang. M. (.<<:). False financial statementsQ characteristics of )hina1s listed companies and )ADT detecting approach, 0nternational Oournal of 0nformation Technology P Decision &a!ing 6 (.) 22;H 24;. *46, Deshmu!h, A., P Talluru, (. (+;;:). A rule9based fuzzy reasoning system for assessing the ris! of management fraud, 0nternational Oournal of 0ntelligent %ystems in Accounting, Finance P &anagement 6 (3) ..2H.3+. *4:, Deshmu!h, A., Domine, O., P %iegel, B.L. (+;;6). &easurement and combination of red flags to assess the ris! of management fraudQ a fuzzy set approach, &anagerial Finance .2 (5) 24H3:. *4;, 'ining, &. &., Oones, D. D., P (oebbec!e, O. A. (+;;6). Deliance on decision aidsQ an examination of auditors8 assessment of management fraud. AuditingQ A Oournal of Bractice and Theory, +5(.), +H+;. *5<, Aotsiantis, %., Aoumana!os, '., Tzelepis, D. P Tampa!as, E. (.<<5). Forecasting fraudulent financial statements using data mining, 0nternational Oournal of )omputational 0ntelligence 2 (.) +<3H++<. *5+, -"usu9Ansah, %., &oyes, F.D. , -yelere, B.$., Lay, B. (.<<.). An empirical analysis of the li!elihood of detecting fraud in ?e" Kealand, &anagerial Auditing Oournal +6 (3) +;.H.<3. *5., Nuan, O., Nuan, )., Deng, N., P Nuan, ). (.<<:). The effects of manager compensation and mar!et competition on financial fraud in public companiesQ an empirical study in )hina, 0nternational Oournal of &anagement .4 (.) (.<<:) 2..H224. *52, )hai, W., Loogs, $.A., P Eerschueren, $.T. (.<<5). Fuzzy Dan!ing of Financial %tatements for Fraud detection. 0n proceeding of 0nternational )onference on Fuzzy %ystem, (.<<5), +4.H+4:. *53, Aiehl, T. D., Loogs, $. A., P (a)omb, ). A. (.<<4). ' ol ing &ulti9Eariate Time9%eries Batterns for the Discrimination of Fraudulent Financial Filings. 0n Broc. of Fenetic and ' olutionary )omputation )onference, .<<4. *54, Aos!i aara, '. (.<<<). Different pre9processing models for financial accounts "hen using neural net"or!s for auditing, Broceedings of the :th 'uropean )onference on 0nformation %ystems, ol. +, .<<<, pp. 2.5H22.:, Eienna, Austria. *55, Feroz, '.L., A"on, T.&., Bastena, E., P Bar!, A.O. (.<<<). The efficacy of red flags in predicting the %')1s targetsQ an artificial neural net"or!s approach, 0nternational Oournal of 0ntelligent %ystems in Accounting, Finance, and &anagement ; (2) +34H+46. *56, Bacheco, D., &artins, A., $arcia, D.&., P Ahator, %. (+;;5). A hybrid intelligent system applied to financial statement analysis, Broceedings of the 4th 0''' conference on Fuzzy %ystems, ol. ., +;;5, pp. +<<6H +<+.:, ?e" -rleans, (A, >%A. *5:, (iou, F. &. (.<<:). Fraudulent financial reporting detection and business failure prediction modelsQ a comparison. &anagerial Auditing Oournal Eol. .2 ?o. 6, pp. 54<955.. *5;, Aapardis, &. A., )hristodoulou, ). P Agathocleous, &. (.<+<). ?eural net"or!sQ the panacea in fraud detectionU &anagerial Auditing Oournal, .4, 54;956:. *6<, Da isan!ar B., Da i E., Dao F.D. P $ose 0. (.<++). Detection of financial statement fraud and feature selection using data mining techni#ues. Decision %upport %ystem, 4<, 3;+94<<. *6+, Berols, O. (.<++). Financial %tatement Fraud DetectionQ An Analysis of %tatistical and &achine (earning Algorithms, AuditingQ A Oournal of Bractice and Theory, Eol. 2<(.), pp. +;94<. *6., Khou, W. and Aapoor, F. (.<++). Detecting e olutionary financial statement fraud. Decision %upport %ystems, . 4<, n. 2, p. 46<9464.

36