International Journal of Computer Applications (0975 – 8887)

Volume 39– No.1, February 2012

A Review of Financial Accounting Fraud Detection

based on Data Mining Techniques

Anuj Sharma Prabin Kumar Panigrahi

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

ABSTRACT financial statement, especially investors and creditors‖ [5].

With an upsurge in financial accounting fraud in the current The accounting fraud is executed by making falsified financial
economic scenario experienced, financial accounting fraud accounting statements where the numbers are manipulated by
detection (FAFD) has become an emerging topic of great overstating assets, spurious entries related to sales and profit,
importance for academic, research and industries. The failure misappropriation in taxes, or understating liabilities, debts,
of internal auditing system of the organization in identifying expenses or losses [1]. The accounting fraud is also defined
the accounting frauds has lead to use of specialized by accounting professionals as ―deliberate and improper
procedures to detect financial accounting fraud, collective manipulation of the recording of data in financial statements
known as forensic accounting. Data mining techniques are in order to achieve an operating profit of the company and
providing great aid in financial accounting fraud detection, appear better than it actually is‖ [6].
since dealing with the large data volumes and complexities of Economically, financial fraud is becoming an increasingly
financial data are big challenges for forensic accounting. This serious problem and effective detecting accounting fraud has
paper presents a comprehensive review of the literature on the always been an important but complex task for accounting
application of data mining techniques for the detection of professionals [7]. The internal auditing of financial matters in
financial accounting fraud and proposes a framework for data the companies has become an increasingly demanding activity
mining techniques based accounting fraud detection. The and there are many evidence that ‗book cooking‘ accounting
systematic and comprehensive literature review of the data practices are world-wide applied for doing financial frauds
mining techniques applicable to financial accounting fraud [8]. The detection of accounting fraud using traditional
detection may provide a foundation to future research in this internal audit procedures is a difficult or sometimes an
field. The findings of this review show that data mining impossible task [9]. First, the auditors usually lack the
techniques like logistic models, neural networks, Bayesian required knowledge concerning the characteristics of
belief network, and decision trees have been applied most accounting fraud. Second, as the fraudulent manipulation of
extensively to provide primary solutions to the problems accounting data is so infrequent, most of the auditors lack the
inherent in the detection and classification of fraudulent data. experience and expertise needed to detect and prevent frauds.
Finally, the other concern people of finance department like
General Terms Chief Financial Officer (CFO), financial managers and
Fraud Detection, Financial Fraud, Financial Statements. accountants are intentionally trying to deceive the internal or
external auditors [10]. While knowing the limitations of an
Keywords audit, finance and accounting managers have concluded that
Financial Accounting Fraud, Fraud Detection, Data Mining. traditional and standard auditing procedures are insufficient to
detect frauds. These limitations of financial auditing suggest
1. INTRODUCTION the need for additional automatic data analysis procedures and
With an upsurge in financial accounting fraud in the current
tools for the effective detection of falsified financial
economic scenario experienced, financial accounting fraud
statements.
detection (FAFD) have received considerable attention from
the investors, academic researchers, media, the financial Although the latest revision auditing standards is enlarging
community and regulators. Due to some high profile financial auditors‘ fraud detection responsibility, effective detecting
frauds discovered and reported at large companies like Enron, accounting fraud has always been a problem for accounting
Lucent, WorldCom and Satyam over the last decade, the profession [5]. The failure of internal auditing system of the
requirement of detecting, defining and reporting financial organization in identifying the accounting frauds has lead to
accounting fraud has increased [1]. use of specialized procedures to detect financial accounting
fraud collective known as forensic accounting. Forensic
The Oxford English Dictionary [2] defines fraud as ―wrongful
accounting plays a vital role in detecting these frauds which
or criminal deception intended to result in financial or
are difficult to find out in internal auditing by employing
personal gain.‖ In academic literature fraud is defined as
accounting, auditing, and investigative skills [11, 12].
leading to the abuse of a profit organization's system without
necessarily leading to direct legal consequences [3]. Although Without the fail-safe financial fraud prevention tools and
the literature is missing a universally accepted definition of procedures, the accounting fraud has become a business
financial fraud, researcher has defined it as ―a deliberate act critical problem in current competitive environment. Data
that is contrary to law, rule, or policy with intent to obtain Mining based financial fraud detection and fraud control,
unauthorized financial benefit‖ [4] and ―intentional automates the whole process and helps to reduce the manual
misstatements or omission of amount by deceiving users of work of screening and checking various statements. This area

37
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

has become one of the key applications of data mining 2. CLASSIFICATION OF DATA
techniques in established industry or government
organizations [3]. MINING TECHNIQUES FOR FRAUD
DETECTION
Data mining is known as gaining insights and identifying In this section, a graphical conceptual framework is proposed
interesting patterns from the data stored in large databases in for the available literature on the applications of data mining
such a way that the patterns and insights are statistically techniques to financial accounting fraud detection. The
reliable, previously unknown, and actionable [13]. Data classification framework, which is shown in Fig. 1, is based
mining is also define as ―a process that uses statistical, on a literature review of existing knowledge on the nature of
mathematical, artificial intelligence and machine learning data mining research [19,20], fraud detection research
techniques to extract and identify useful information and [1,3,16,17,18].
subsequently gaining knowledge from a large database‖ [14,
15]. The blending point between data mining and detecting A classification framework for financial fraud is suggested in
accounting fraud is that, data mining as an advanced [7] based on the financial crime framework of the U.S.
analytical tool may assist the auditors in decision making and Federal Bureau of Investigation [21], which is one of the
detecting fraud. The data mining techniques have the potential established frameworks for financial fraud Detection. Fig. 1
to solve the contradiction between effect and efficiency of consists of two layers, the first comprising the six data mining
fraud detection [5]. Data mining plays an important role in application classes of classification, clustering, prediction,
financial accounting fraud detection, as it is often applied to outlier detection, regression, and visualization
extract and discover the hidden patterns in very large [3,16,18,22,23], supported by a set of algorithmic approaches
collection of data [7]. An auditor can never become certain to extract the relevant relationships in the data [14].
about the legitimacy of and intention behind a fraudulent
transaction. Concerning this reality, the most optimal and cost
effective option is to find out enough evidences of fraud from
the available data using specialized mathematical and data
processing algorithms [3]. 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
techniques for detection of financial accounting fraud is a
promising field, many review articles have published in
conference proceedings or journal publications. Statistical
methods of detecting different types of frauds like credit card
fraud, fraudulent money laundering, telecommunications
fraud, etc., are reviewed in [16]. Applications of data mining
in stock markets and bankruptcy predictions and related fraud
detection have been surveyed in [17]. A survey of data
mining-based fraud detection research is presented in [3], Fig. 1: The Conceptual Framework for Application of
including credit transaction fraud, telecoms subscription Data Mining to FAFD
fraud, automobile insurance fraud, terrorist detection, A brief description of the conceptual framework with
financial crime detection, intrusion and spam detection. references is provided and of the six data mining application
Others researcher have reviewed insurance fraud [18] and classes (classification, clustering, outlier detection, prediction,
financial statement fraud [1]. regression and visualization), each component of is discussed
This paper presents a comprehensive review of the research in more detail in the following sections.
literature on the application of data mining techniques to
detect financial accounting fraud. The paper proposes a
2.1 Classification of Data Mining
framework for data mining techniques based accounting fraud Applications
detection to help certified public accountants selecting
Each of the six data mining application classes is supported by
suitable data and data mining technologies for detecting fraud.
a set of algorithmic approaches to extract the relevant
The systematic and comprehensive literature review of the
relationships in the data. These approaches can handle
data mining techniques applicable to financial accounting
different classes of problems. The classes are presented
fraud detection may provide a foundation to future research in
below.
this field.
Classification - Classification builds up (from the training
The rest of the paper is organized as follows. Section 2
set) and utilizes a model (on the target set) to predict the
describes classification of data mining techniques and
categorical labels of unknown objects to distinguish between
applications for financial accounting fraud detection. Section
objects of different classes. These categorical labels are
3 provides distribution of the research literature as per the
predefined, discrete and unordered [24]. The research
applications and techniques of data mining for the detection of
literature describes that classification or prediction is the
financial accounting fraud. Section 4 describes our framework
process of identifying a set of common features (patterns), and
in more detail. Section 5 outlines future direction of this work.
proposing models that describe and distinguish data classes or
concepts [17]. Common classification techniques include
neural networks, the Naïve Bayes technique, decision trees
and support vector machines. Such classification tasks are
used in the detection of credit card, healthcare and automobile
insurance, and corporate fraud, among other types of fraud,

38
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

and classification is one of the most common learning models techniques are logistic models, neural networks, the Bayesian
in the application of data mining in fraud detection. belief network, and decision trees, all of which fall into the
―classification‖ category. These four techniques are discussed
Clustering - Clustering is used to partition objects into in more detail in the following paragraphs.
previously unknown conceptually meaningful groups (i.e.
clusters), with the objects in a cluster being similar to one Regression Models - The regression based models are mostly
another but very dissimilar to the objects in other clusters. used in financial accounting fraud detection. The majority of
Clustering is also known as data segmentation or partitioning them are based on logistic regression, stepwise-logistic
and is regarded as a variant of unsupervised classification regression, multi criteria decision making method and
[24]. Cluster analysis decomposes or partitions a data set exponential generalized beta two (EGB2) [7]. Logistic model
(single or multivariate) into dissimilar groups so that the data is a generalized linear model that is used for binomial
points in one group are similar to each other and are as regression in which the predictor variables can be either
different as possible from the data points in other groups [1]. numerical or categorical [30, 31]. It is principally used to
It is suggested that data objects in each cluster should have solve problems caused by insurance and corporate fraud.
high intra-cluster similarity within the same cluster but should
have low inter-cluster similarity to those in other clusters [17]. Some of the research has suggested logistic regression based
The most common clustering techniques are the K-nearest model to predict the presence of financial statement fraud [30,
neighbour, the Naïve Bayes technique and self-organizing 33]. Statistical method of logistic regression can detect
maps. falsified financial statements efficiently [30]. Some
researchers have also developed generalized qualitative
Prediction - Prediction estimates numeric and ordered future response model based on Probit and Logit techniques to
values based on the patterns of a data set [19]. It is noted that, predict financial statement fraud. That model was based on a
for prediction, the attribute, for which the value being dataset collected by an international public accounting
predicted is continuous-valued (ordered) rather than company and needs testing for generalization [34]. Cascaded
categorical (discrete-valued and unordered). This attribute is Logit model has also proposed to investigate the relationship
referred as the predicted attribute [24]. Neural networks and between insider trading and possibility of fraud. The study in
logistic model prediction are the most commonly used [35] found that, when the fraud is being executed, insiders, i.e.
prediction techniques. top executives and managers, reduce their stock holdings
through high stock selling activity. The other methods like
Outlier Detection - Outlier detection is employed to measure statistical regression analysis are also useful to test if the
the ―distance‖ between data objects to detect those objects existence of an independent audit committee mitigates or
that are grossly different from or inconsistent with the reduces the likelihood of fraud. Literature also describes that
remaining data set [24]. Data that appear to have different organizations with audit committees, formed by independent
characteristics than the rest of the population are called managers, meeting no more than twice per year, are less likely
outliers [26]. The problem of outlier/anomaly detection is one to be sanctioned for fraudulent financial reporting [36].
of the most fundamental issues in data mining. A commonly
used technique in outlier detection is the discounting learning In 2000, Bell and Carcello [37] proposed a logistic regression
algorithm [27]. model for estimating the likelihood of fraudulent financial
reporting for an audit client. The model was conditioned on
Regression - Regression is a statistical methodology used to the presence or absence of several fraud-risk factors. The
reveal the relationship between one or more independent fraud risk factors identified in the final model included weak
variables and a dependent variable (that is continuous-valued) internal control system, rapid company growth, inadequate or
[24]. Many empirical studies have used logistic regression as inconsistent relative profitability, management that just want
a benchmark [28]. The regression technique is typically to achieve earnings projections anyhow while lying to the
undertaken using such mathematical methods as logistic auditors or is overly evasive, company ownership status
regression and linear regression, and it is used in the detection (public vs. private), and interaction term between a weak
of credit card, crop and automobile insurance, and corporate control environment and an aggressive management attitude
fraud. towards financial reporting.
Visualization - Visualization refers to the easily In 2002, Spathis et al. [38] proposed that statistical techniques
understandable presentation of data and to methodology that like logistic regression may be suitable to develop a model to
converts complicated data characteristics into clear patterns to identify factors related to fraudulent financial statement. Non-
allow users to view the complex patterns or relationships parametric regression-based framework was used to run the
uncovered in the data mining process [14]. The researchers falsified financial statement detection model. The proposed
have exploited the pattern detection capabilities of the human model was compared with discriminant analysis and logit
visual system by building a suite of tools and applications that regression methods for benchmarking.
flexibly encode data using colour, position, size and other
visual characteristics. Visualization is best used to deliver The regression analysis using Logit model can be used for
complex patterns through the clear presentation of data or empirical analysis of financial indexes which can significantly
functions [29]. predict financial fraud [39]. Logistic analysis and clustering
analysis jointly can be used to establish a detecting model of
2.2 Classification of Data Mining fraud from four aspects of financial indexes, company
Techniques for Financial Accounting Fraud governance, financial risk and pressure and related trading.
After cluster filtering significant variables, prediction model
Detection can be established with methods of Standardization, non-
To determine the main algorithms used for financial Standardization Bayes and Logistic [40].
accounting fraud detection, we present a Review of data The logistic regression based accounting fraud detecting
mining techniques identified in literature applied to the models are common in literature since the model based on
detection of financial fraud. The most frequently used

39
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

logistic regression can reach up to 95.1% of detecting Naïve Bayes - Naïve Bayes is used as simple probabilistic
accuracy with significant expectation effect [41]. classifier based on Bayes conditional probability rule. Naïve
Bayes follows strong (naive) statistical independence
Neural Networks – The neural networks are non- assumptions for the predictor variables. It is an effective
linear statistical data modeling tools that are inspired by the classification tool that is easy to interpret and particularly
functionality of the human brain using a set of interconnected suited when the dimensionality of the inputs is high [24]. In a
nodes [31, 32]. Neural networks are widely applied in study [3], Naïve Bayes classifier outperformed the
classification and clustering, and its advantages are as follows. conventional classifier. The efficiency of predicting financial
First, it is adaptive; second, it can generate robust models; and fraud data was higher with no false positives with relatively
third, the classification process can be modified if new low false negatives. Naïve Bayes methods are widely used in
training weights are set. Neural networks are chiefly applied banking and financial fraud detection and claim fraud
to credit card, automobile insurance and corporate fraud. detection. To apply Ada Boosted Naïve Bayes scoring to
Literature describes that neural networks can be used as a insurance claims fraud, a case study is given to diagnosis
financial fraud detection tool. The neural network fraud claim fraud [45].
classification model employing endogenous financial data Nearest Neighbour Method - Nearest neighbour method is a
created from the learned behavior pattern can be applied to a similarity based classification approach. Based on a
test sample [42]. The neural networks can be used to predict combination of the classes of the most similar k record(s),
the occurrence of corporate fraud at the management level every record is classified. Sometimes this method is also
[43]. known as the k-nearest neighbour technique [24]. K-nearest
Researchers have explored the effectiveness of neural neighbour method is used in automobile insurance claims
networks, decision trees and Bayesian belief networks in fraud detection [46] and for identifying defaults of credit card
detecting fraudulent financial statements (FFS) and to identify clients [31].
factors associated with FFS [8]. Fuzzy logic and Genetic Algorithm – Genetic algorithms are
The study in [10] revealed that input vector consisted of used in classifier systems to represent and modeling the
financial ratios and qualitative variables, was more effective auditor decision behavior in a fraud setting [47]. Genetic
when fraud detection model was developed using neural algorithm along with binary support vector system (BSVS)
network. The model was also compared with standard which is based on the support vectors in support vector
statistical methods like linear and quadratic discriminant machines (SVM) are used to solve problems of credit card
analysis, as well as logistic regression methods [10]. fraud that had not been well identified [48].

The generalized adaptive neural network architectures and the Fuzzy Logic is a mathematical technique that classifies
adaptive logic network are well received for fraud detection. subjective reasoning and assigns data to a particular group, or
The hybrid techniques like fuzzy rule integrated with a neural cluster, based on the degree of possibility the data has of
network (neuro-fuzzy systems) are also proposed. The being in that group. The expert fuzzy classification techniques
literature describes that the integrated fuzzy neural network enable one to perform approximate reasoning that can
outperformed traditional statistical models and neural improve performance in three ways. First, performance is
networks models reported in prior studies [44]. improved through efficient numerical representation of vague
terms, because the fuzzy technology can numerically show
Bayesian Belief Network - The Bayesian belief network representation of a data item in a particular category. The
(BBN) represents a set of random variables and their second way performance is enhanced is through increased
conditional independencies using a directed acyclic graph range of operation in ill-defined environments, which is the
(DAG), in which nodes represent random variables and way that fuzzy methodology can show partial membership of
missing edges encode conditional independencies between the data elements in one or more categories that may not be
variables [8]. The Bayesian belief network is used in clearly defined in traditional analysis. Finally, performance is
developing models for credit card, automobile insurance, and increased because the fuzzy technology has decreased
corporate fraud detection. The research in [8] described that sensitivity to "noisy" data, or outliers. A multilevel fuzzy rule-
Bayesian belief network model correctly classified 90.3% of based system is proposed in [49] to rank state financial
the validation sample for fraud detection. Bayesian belief management. The authors used fuzzy set theory to represent
network outperformed neural network and decision tree imprecision in evaluated information and judgments.
methods and achieved outstanding classification accuracy [8].
A fuzzy logic model has been implemented in [50] for fraud
Decision Trees – A decision tree (DT) is a tree structured detection in an Excel spreadsheet. By using the fuzzy logic
decision support tool, where each node represents a test on an model to develop clusters for different statements representing
attribute and each branch represents possible consequences. In red flags in the detection of fraud, non-financial data was
this way, the predictive model attempts to divide observations included with financial statement variables for the analysis.
into mutually exclusive subgroups and is used for data mining The model consist of different financial variables like
and machine learning tasks [8]. Decision trees are predictive leverage, profitability, liquidity, cash flow and a variable
decision support tools that create mapping from observations designed to represent a company‘s risk of fraud. Fuzzy logic
to possible consequences [24]. These trees can be planted via efficiently modeled the variable, which was developed to
machine-learning-based algorithms such as the ID3, CART quantify fraud risk factors. The model predicted frauds with
and C4.5. Predictions are represented by leaves and the 86.7% accuracy [50]. The same model was adapted in [51] to
conjunctions of features by branches. Decision trees are develop a model for detection of financial statement fraud.
commonly used in credit card, automobile insurance, and The proposed model used a combination of different financial
corporate fraud. To identify and predict the impact of statement data. The data included in the model consisted of
fraudulent financial statements, classification and regression easily available non-financial information, and a specialized
trees (CART) algorithm is introduced in [56]. variable constructed by quantifying the company‘s financial
statements issues using fuzzy logic. The study revealed that

40
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

most successful predictor variables were financial variables expert system and genetic algorithm for financial accounting
indicating cash flow, cash liquidity, the ratio of sales to assets, fraud detection.
and a variable indicating company size. The model predicted
frauds with 76.7 % accuracy [51]. 4. DATA MINING BASED
Fuzzy logic based expert system has been developed to FRAMEWORK FOR FRAUD
identify and evaluate whether elements of fraud are involved DETECTION
in insurance claims settlements. The fuzzy logic based expert The research related with application of data mining
system was developed for auditors to identify fraud in settled algorithms and techniques for financial accounting fraud
claimed insurance. The system was able to cut costs by detection is a well studied area. The implementation of these
detecting fraudulent filings [52]. The other fuzzy logic based techniques follows the same information flow of data mining
fraud detection systems are given in [57, 58]. processes in general. The process starts with feature selection
then proceeds with representation, data collection and
Genetic programming with fuzzy logic production rules is management, pre-processing, data mining, post-processing,
used to classifying data. The study in [53] has proposed and and in the end performance evaluation. This paper has
tested a system to detect frauds on real home insurance claims proposed an expanded generic data mining framework. This
and credit card transaction data. The study on genetic framework considers specific characteristics of fraud
programming for fraud detection lacks benchmarking with the detection techniques for financial accounting fraud (see Fig.
existing methods and techniques. A genetic algorithm based 2).
approach to detect financial statement fraud in presented in
[54]. It was found that exceptional anomaly scores are 5. CONCLUSION AND FUTURE
valuable metrics for characterizing corporate financial
behavior and that analyzing these scores over time represents RESEARCH
an effective way of detecting potentially fraudulent behavior. This paper reviewed the literature describing use of data
Another study on fraud detection applied many different mining algorithms including statistical test, regression
classification techniques in a supervised two-class setting and analysis, Neural Network, decision tree, Bayesian network etc
a semi-supervised one-class setting in order to compare the for financial accounting fraud detection. Regression Analysis
performances of these techniques and settings [55]. is widely used for fraud detection since it has great
explanation ability. Different regression model used by
Expert Systems – Researchers in the field of Expert systems researchers are Logit, Step-wise Logistic, UTADIS and EGB2
have examined the role of Expert Systems in increasing the etc. Neural Networks are important tool for data mining. The
detecting ability of auditors and statement users. By using researchers have not made any comparison so far, related with
expert system, they could have better detecting abilities to detecting effect and accuracy of Neural Network compared to
accounting fraud risk under different context and level and regression model. The advantages of Neural Network are that
enable auditors give much reliable auditing suggestions there are no strict requests for data and it has a strong
through rational auditing procedure. The research has generalization and adjustment. After correct allocation and
confirmed that the use of an expert system enhanced the proper training, Neural Network may perform great
auditors‘ performance. With assistance from expert system, classification comparing with regression model. But due to
the auditors discriminated better, among situations with special inner hidden structure, it is impossible for researchers
different levels of management fraud-risk. Expert System to track the formation process of output conclusion. There are
aided in decision making regarding appropriate audit actions other issues also related with Neural Network like no clear
[59]. explanation on connecting weight, complex accuracy and
statistical reliability checking procedure, and lack of
The financial accounting fraud detection research is classified
explanation.
as per data mining application and data mining techniques.
Some researchers have tried to apply a combination of many This paper suggests that using only financial statements data
data mining techniques like decision trees, neural networks, may not be sufficient for detections of fraud. The importance
Bayesian belief network, K-nearest neighbour as in [60]. The of data mining techniques in the detection of financial fraud
main objective is to apply a hybrid decision support system has been recognized. The future work may be proposing a
using stacking variant methodology to detect fraudulent comprehensive classification framework or a systematic
financial statements. Some research is targeted to indentify review of data mining application in financial accounting
financial frauds using simple logistic regression models in fraud detection. In this study, we conduct an extensive review
specific country like in New Zealand [61] and China [62]. of academic articles and provide a comprehensive
bibliography and classification framework for the applications
Some of the recent study in the financial fraud detection make
of data mining to Financial Accounting Fraud Detection. The
used of two or more data mining applications in a hybrid
intention is to inform both academics and practitioners of the
manner or just attempt to compare their effectiveness [70-72].
areas in which specific data mining techniques can be applied
3. THE RESEARCH UNDER PROPOSED to financial accounting fraud detection, and to report and
compile a systematic review of the burgeoning literature on
CLASSIFICATION FRAMEWORK financial accounting fraud detection. Although our study
The Table 1, 2, 3 and 4 show the distribution of the research cannot claim to be exhaustive, we believe that it will prove a
papers as per the order of the year of publication. useful resource for anyone interested in financial accounting
The Table 1 presents the research work on neural network for fraud detection research, and will help simulate further
financial accounting fraud detection. The literature review is interest in the field.
presented in the order of publication. The Table 2 presents
the research work on regression models for financial
accounting fraud detection. The Table 3 shows the research
work on fuzzy logic and Table 4 presents research work on

41
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

Table 1: Research on Neural Network for Financial Accounting Fraud Detection

S. Data Mining
Author Main Objective Reference
No. Techniques
Fanning, Cogger and To use neural networks to develop a model for
1 Neural Networks [9]
Srivastava (1995) detecting managerial fraud
To develop a neural network fraud classification
Green and Choi
2 Neural Networks model employing endogenous financial data in [42]
(1997)
corporate fraud
Fanning and To use neural networks to develop a model for
3 Neural Networks [10]
Cogger (1998) detecting managerial fraud
Cerullo and Cerullo To use neural networks to predict the occurrence of
4 Neural Networks [43]
(1999) corporate fraud at the management level
To investigate the impact of various pre-processing
5 Koskivaara (2000) Neural Networks models on the forecast capability of neural network [65]
for auditing financial accounts
To predict the possible fraudsters and accounting
6 Feroz et al. (2000) Neural Networks [66]
manipulations

Lin, Hwang and Fuzzy Neural Network, To evaluate the utility of an integrated fuzzy neural
7 [44]
Becker (2003) Logistic Model network model for corporate fraud detection
Decision Trees, Neural
To apply a hybrid decision support system using
Kotsiantis et al. Networks, Bayesian
8 stacking variant methodology to detect fraudulent [60]
(2006) Belief Network, K-
financial statements
Nearest Neighbour
Neural Networks, To explore the effectiveness of neural networks,
Kirkos et al. Decision Trees, decision trees and Bayesian belief networks in
9 [8]
(2007) Bayesian Belief detecting fraudulent financial statements (FFS)
Network and to identify factors associated with FFS
Fen-May Liou Neural Networks To build detection/prediction models for detecting
10 [68]
(2008) fraudulent financial reporting
M Krambia-
To test the use of artificial neural networks as a
11 Kapardis et al. Neural Networks [69]
tool in fraud detection
(2010)
Neural Network,
Ravisankar et al. To identify companies that resort to financial
12 Support Vector [70]
(2011) statement fraud
Machines
Neural Networks, To compares the performance of popular statistical
13 Perols (2011) Support Vector and machine learning models in detecting financial [71]
Machines statement fraud
To detect financial statement fraud with exploring
Zhou and Kapoor Neural Networks,
14 a self-adaptive framework (based on a response [72]
(2011) Bayesian Networks
surface model) with domain knowledge

Table 2: Research on Regression Models for Financial Accounting Fraud Detection

Data Mining
S. No. Author Main Objective Reference
Techniques

1 Persons (1995) Logistic Model To detect financial reporting frauds [25]

Logit Regression To predict the presence of financial statement
2 Beasley (1996) [33]
Analysis fraud
Probit And Logit
3 Hansen et al. (1996) To use Probit and Logit techniques to predict fraud [34]
Techniques
Summers and Logit Regression To investigate the relationship between insider
5 [35]
Sweeney (1998) Analysis trading and fraud
Abbot, Park and Statistical Regression To examine if the existence of an independent
4 [36]
Parker (2000) Analysis audit committee mitigates the likelihood of fraud
Bell and Carcello To develop a logistic regression model to estimate
6 Logistic Model [37]
(2000) fraudulent financial reporting for an audit client

42
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

To use logistic regression to examine published

7 Spathis (2002) Logistic Regression data and develop a model to detect the factors [30]
associated with FFS
Spathis, Doumpos,
To develop a model to identify factors associated
8 and Zopounidis Logistic Regression [38]
with fraudulent financial statement
(2002)
Owusu-Ansah et al. Logistic Regression To explore the Logit regression model to detect
9 [61]
(2002) Models corporate fraud in New Zealand
Xuemin Huang Regression Analysis To analyze financial indexes which can predict
10 [39]
(2006) Using Logit Model financial fraud
Logistic Analysis And To establish a detecting model of fraud which can
11 Haisong Ren (2006) [40]
Clustering Analysis be used for empirical analysis of financial indexes
12 Guoxin et al. (2007) Logistic Regression To develop accounting fraud detecting model [41]
Classification And To introduce classification and regression trees to
Bai, Yen and Yang
13 Regression Trees identify and predict the impact of fraudulent [56]
(2008)
(CART) financial statements
To employ a logistic regression model to test the
Logistic Regression effects of managerial compensation and market
14 Yuan et al. (2008) [62]
Models competition on financial fraud among listed
companies in China
Logistic Regression, To build detection/prediction models for detecting
15 Fen-May Liou (2008) [68]
Classification Trees fraudulent financial reporting

Ravisankar et al. To identify companies that resort to financial

16 Logistic Regression [70]
(2011) statement fraud
To compares the performance of popular statistical
Logistic Regression,
17 Perols (2011) and machine learning models in detecting financial [71]
C4.5
statement fraud
Zhou and Kapoor Regression, Decision
18 To detect financial statement fraud [72]
(2011) Tree

Table 3: Research on Fuzzy Logic for Financial Accounting Fraud Detection

S. Data Mining
Author Main Objective Reference
No. Techniques
Deshmukh, Romine To provide a fuzzy sets model to assess the risk
1 Fuzzy Logic [58]
and Siegel (1997) of managerial fraud
Deshmukh and Rule-Based Fuzzy To build a rule-based fuzzy reasoning system to
2 [57]
Talluru (1998) Reasoning System assess the risk of managerial fraud
Ammar, Wright and To use fuzzy set theory to represent imprecision
3 Fuzzy Logic [49]
Selden (2000) in evaluated information and judgments
To develop fuzzy logic model to develop clusters
Lenard and Alam Fuzzy Logic and
4 for different statements representing red flags in [50]
(2004) Expert Reasoning
the detection of fraud
Pathak, Vidyarthi
Fuzzy Logic and
5 and Summers To identify fraud in settled claims [52]
Expert System
(2005)
To convert binary classification rules learned
Chai, Hoogs and
6 Fuzzy Logic from a genetic Algorithm to a fuzzy score for [63]
Verschueren (2006)
financial data fraud rule matching
Lenard, Watkins To detect financial statement fraud using fuzzy
7 Fuzzy Logic [51]
and Alam (2007) logic

Table 4: Research on Expert System and Genetic Algorithm for Financial Accounting Fraud Detection

S. Data Mining
Author Main Objective Reference
No. Techniques
Hybrid intelligent
Pacheco et al.
1 system with NN and To diagnose financial problems in companies [67]
(1996)
fuzzy expert system
Eining, Jones and To build an expert system applying the Logit
2 Expert System [59]
Loebbecke (1997) statistical model to enhance user engagement and

43
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

increase reliance on the aid

To use genetic algorithms to aid the decisions of

Evolutionary
Welch, Reeves and Defense Contractor Audit Agency (DCAA)
3 algorithms (genetic [47]
Welch (1998) auditors when they are estimating the likelihood
algorithms)
of contracts fraud
Kiehl, Hoogs and Genetic Algorithm To automatically detect financial statement fraud
4 [64]
LaComb (2005)
Genetic Algorithm To detect financial statement fraud based on
Hoogs et al.
5 anomaly scores as a metrics for characterizing [54]
(2007)
corporate financial behavior
Supervised and Semi- To detect financial statement fraud
Juszczak et al.
6 Supervised [55]
(2008)
Classification

Feature Selection and Representation

Data Sources
Financial Statements
Audit Reports
Internal Control Data Cleaning and Integration
Account Ledgers
Training/ Test Data Set
Fraud Firm
Non-Fraud Firms
Data Selection and Transformation

DM Application DM Techniques
Classification
Clustering Regression
Data Mining (DM) Neural Networks
Prediction
Outlier detection Bayesian Belief Network
Regression Naïve Bayes
Visualization Nearest Neighbour
Pattern Evaluation Fuzzy logic
Decision Tree
Genetic Algorithm
Expert System
Post Processing of Patterns and Trends

Metrics
Performance Evaluation Error Rate

Fig. 2: The Data Mining Based Framework for Financial Accounting Fraud Detection

44
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

The data mining techniques of outlier detection and [2] Oxford Concise English Dictionary, 11th Edition, Oxford
visualization have seen only limited use. The lack of research University Press, 2009.
on the application of outlier detection techniques to financial [3] Phua, C., Lee, V., Smith, K. & Gayler, R. (2005). A
accounting fraud detection may be due to the difficulty of comprehensive survey of data mining-based fraud
detecting outliers. Indeed, research has point out that outlier detection research, Artificial Intelligence Review (2005)
detection is a very complex task akin to finding a needle in a 1–14.
haystack. Distinct from other data mining techniques, outlier
detection techniques are dedicated to finding rare patterns [4] Wang, J., Liao, Y., Tsai, T. & Hung, G. (2006).
associated with very few data objects. In the field of financial Technology-based financial frauds in Taiwan: issue and
accounting fraud detection, outlier detection is highly suitable approaches, IEEE Conference on: Systems, Man and
for distinguishing fraudulent data from authentic data, and Cyberspace Oct (2006) 1120–1124.
thus deserves more investigation. Similarly, visualization [5] Wang, S. (2010). A Comprehensive Survey of Data
techniques have a strong ability to recognize and present data Mining-Based Accounting-Fraud Detection Research.
anomalies, which could make the identification and International Conference on Intelligent Computation
quantification of fraud schemes much easier. Technology and Automation, vol. 1, pp.50-53, 2010.
The paper suggests that one of the reasons for the limited [6] Accounting Fraud Definition and Examples retrieved
number of relevant journal articles published for financial from http://www.accountingelite.com/accounting-
accounting fraud detection is the difficulty of obtaining tips/accounting-fraud-definition-and-examples-free-
sufficient research data. Fanning and Cogger [9] highlight the accounting-fraud-article/
challenge of obtaining fraudulent financial statements, and [7] Ngai, E.W.T., Hu, Y., Wong, Y. H., Chen, Y., & Sun, X.
note that this creates enormous obstacles in financial (2010). The application of data mining techniques in
accounting fraud detection research. The most urgent financial fraud detection: A classification framework and
challenge facing financial accounting fraud detection is to an academic review of literature, Decision Support
bridge the gap between practitioners and researchers. The System (2010), doi:10.1016/j.dss.2010.08.006.
existing financial accounting fraud detection research
concentrates on particular types of data mining techniques or [8] Kirkos, E., Spathis, C., & Manolopoulos, Y. (2007).
models, but future research should direct its attention toward Data mining techniques for the detection of fraudulent
finding more practical principles and solutions for financial statements, Expert Systems with Applications
practitioners to help them to design, develop, and implement 32 (4) (2007) 995–1003.
data mining and business intelligence systems that can be [9] Fanning, K., Cogger, K., & Srivastava, R. (1995).
applied to financial accounting fraud detection. Detection of management fraud: a neural network
approach. International Journal of Intelligent Systems in
We predict that increasing amounts of privacy-preserving
Accounting, Finance & Management, vol. 4, no. 2, pp.
financial data will be publicly available in the near future due
113– 26, June 1995.
to increased collaboration between practitioners and
researchers, and that this should lead to more investigations of [10] Fanning, K., & Cogger, K. (1998). Neural network
data mining techniques that can be applied to privacy- detection of management fraud using published financial
preserving data. data. International Journal of Intelligent Systems in
Accounting, Finance & Management, vol. 7, no. 1, pp.
A further problem faced by financial accounting fraud 21- 24, 1998.
detection is that of cost sensitivity. The cost of
misclassification (false positive and false negative errors) [11] Silverstone, Howard, & Sheetz, M. (2004). Forensic
differs, with a false negative error (misclassifying a fraudulent Accounting and Fraud Investigation for Non-Experts.
activity as a normal activity) usually being more costly than a Hoboken, John Wiley & Sons, 2004.
false positive error (misclassifying a normal activity as a [12] Bologna, Jack & Lindquist, R. J. (1987). Fraud Auditing
fraudulent activity) [3]. Few studies have explicitly included and Forensic Accounting. New York: John Wiley &
cost in their financial accounting fraud detection modeling Sons, 1987.
[28], but future research on the application of data mining
[13] Elkan, C. (2001). Magical Thinking in Data Mining:
techniques to financial accounting fraud detection problems
Lessons from COIL Challenge 2000. Proc. of
should take into account cost sensitivity considerations.
SIGKDD01, 426-431.
This study has two major limitations. First, our review applied [14] Turban, E., Aronson, J.E., Liang, T.P., & Sharda, R.
several keywords to search only some online databases for (2007). Decision Support and Business Intelligence
articles published between 1992 and 2011. A future review Systems, Eighth edition, Pearson Education, 2007.
could be expanded in scope. Second, we considered only
articles for financial accounting fraud. Future research could [15] Frawley, W. J., Piatetsky-Shapiro, G., & Matheus, C.J.
be expanded to include relevant data mining application for (1992). Knowledge discovery in databases: An
detection of frauds in other area like health, communication, overview, AI Magazine 13 (3) (1992) 57–70.
insurance, banking etc. [16] Bolton, R. J., & Hand, D.J. (2002). Statistical fraud
detection: a review, Statistical Science 17 (3) (2002)
6. REFERENCES 235–255.
[1] Yue, X., Wu, Y., Wang, Y. L., & Chu, C. (2007). A
[17] Zhang, D., & Zhou, L. (2004). Discovering golden
review of data mining-based financial fraud detection
research, international conference on wireless nuggets: data mining in financial application, IEEE
communications Sep, Networking and Mobile Transactions on Systems, Man and Cybernetics 34 (4)
(2004) Nov.
Computing (2007) 5519–5522.
[18] Derrig, R. A. (2002). Insurance fraud, The Journal of
Risk and Insurance 69 (3) (2002) 271–287.

45
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

[19] Ahmed, S.R. (2004). Applications of data mining in [35] Summers, S. L., & Sweeney, J. T. (1998). Fraudulent
retail business, International Conference on Information misstated financial statements and insider trading: an
Technology: Coding and Computing 2 (2) (2004) 455– empirical analysis. The Accounting Review, 73(1), 131–
459. 146.
[20] Mitra, S., Pal, S.K., & Mitra, P. (2002). Data mining in [36] Abbot, J. L., Park, Y., & Parker, S. (2000). The effects of
soft computing framework: a survey, IEEE Transactions audit committee activity and independence on corporate
on Neural Networks 13 (1) (2002) 3–14. fraud. Managerial Finance, 26(11), 55–67.
[21] FBI, Federal Bureau of Investigation, Financial Crimes [37] Bell, T., & Carcello, J. (2000). A decision aid for
Report to the Public Fiscal Year, Department of Justice, assessing the likelihood of fraudulent financial reporting.
United States, 2007, http://www.fbi.gov/publications/ Auditing: A Journal of Practice & Theory, 9(1), 169–
financial /fcs_ report2007/financial_crime_2007.htm. 178.
[22] Fawcett, T., & Provost, F. (1997). Adaptive fraud [38] Spathis, C., Doumpos, M., & Zopounidis, C. (2002).
detection, Data Mining and Knowledge Discovery 1 (3) Detecting falsified financial statements: a comparative
(1997) 291–316. study using multicriteria analysis and multivariate
[23] Sánchez, D., Vila, M.A., Cerda, L., & Serrano, J.M. statistical techniques. The European Accounting Review,
(2009). Association rules applied to credit card fraud 11(3), 509–535.
detection, Expert Systems with Applications 36 (2) [39] Huang, X. (2006). Research on Public Company
(2009) 3630–3640. Accounting Fraud and regulation-from perspective of
[24] Han, J., & Kamber, M. (2006). Data Mining: Concepts protecting investors [D]. Xiamen: Xiamen University,
and Techniques, Second edition, Morgan Kaufmann 2006.
Publishers, 2006, pp. 285–464. [40] Ren, H. (2006). Investigation Research on Public
[25] Persons O.S. (1995). Using Financial Statement Data to company financial report fraud. [D] Dalian: Dongbei
Identify Factors Associated with Fraudulent Financial University of Finance, 2006.
Reporting .Journal of Applied Business [41] Chen, G., Zhanjia L., & Feng, H. (2007). Empirical
Research,1995,11(3):38-46. Study on detecting financial statements Fraud- based on
[26] Agyemang, M., Barker, K., & Alhajj, R. (2006). A empirical data of public companies. [J] Auditing Study,
comprehensive survey of numeric and symbolic outlier 2007.
mining techniques, Intelligent Data Analysis 10 (6) [42] Green, B. P., & Choi, J. H. (1997). Assessing the risk of
(2006) 521–538. management fraud through neural-network technology.
[27] Yamanishi, K., Takeuchi, J., Williams, G., & Milne, P. Auditing: A Journal of Practice and Theory, 16(1), 14–
(2004). On-line unsupervised outlier detection using 28.
finite mixtures with discounting learning algorithms, [43] Cerullo, M. J., Cerullo, V. (1999). Using neural networks
Data Mining and Knowledge Discovery 8 (3) (2004) to predict financial reporting fraud, Computer Fraud &
275–300. Security May/June (1999) 14–17.
[28] Viaene, S., Derrig, R.A., Baesens, B., & Dedene, G. [44] Lin, J. W., Hwang, M. I., & Becker, J. D. (2003). A
(2002). A comparison of state-of-the-art classification Fuzzy Neural Network for Assessing the Risk of
techniques for expert automobile insurance claim fraud Fraudulent Financial Reporting. Managerial Auditing
detection, The Journal of Risk and Insurance 69 (3) Journal, 2003, 18(8):657-665.
(2002) 373–421. [45] Viaene, S., Derrig, R. A., & Dedene G. (2004). A Case
[29] Eick, S.G. & Fyock, D.E. (1996). Visualizing corporate Study of Applying Boosting Naive Bayes to Claim Fraud
data, AT&T Technical Journal 75 (1) (1996) 74–86. Diagnosis, IEEE Transactions on Knowledge and Data
[30] Spathis, C. T. (2002). Detecting false financial Engineering, v.16 n.5, p.612-620, May 2004.
statements using published data: some evidence from [46] Viaene, S., Derrig, R. A., Baesens, & Dedene, G. (2002).
Greece, Managerial Auditing Journal 17 (4) (2002) 179– A comparison of state-of-the-art classification techniques
191. for expert automobile insurance claim fraud detection,
[31] Yeh, I., & Lien, C. (2008). The comparisons of data The Journal of Risk and Insurance 69 (3) (2002) 373–
mining techniques for the predictive accuracy of 421.
probability of default of credit card clients, Expert [47] Welch, J., Reeves, T. E., & Welch, S. T. (1998). Using a
Systems with Applications 36 (2) (2008) 2473–2480. genetic algorithm-based classifier system for modeling
[32] Ghosh, S., & Reilly, D. L. (1994). Credit card fraud auditor decision behavior in a fraud setting, International
detection with a neural-network, 27th Annual Hawaii Journal of Intelligent Systems in Accounting, Finance &
International, Conference on System Science 3 (1994) Management 7 (3) (1998) 173–186.
621–630. [48] Chen, R., Chen, T., & Lin, C. (2006) A new binary
[33] Beasley, M. (1996). An empirical analysis of the relation support vector system for increasing detection rate of
between board of director composition and financial credit card fraud, International Journal of Pattern
statement fraud. The Accounting Review, 71(4), 443– Recognition and Artificial Intelligence 20 (2) (2006)
466. 227–239.

[34] Hansen, J. V., McDonald, J. B., Messier, W. F., & Bell, [49] Ammar, Salwa, Wright, R., & Selden, S. (2000).
T. B. (1996). A generalized qualitative—response model Ranking State Financial Management: A Multilevel
and the analysis of management fraud. Management Fuzzy Rule-based System.‖ Decision Sciences. 31: 449-
Science, 42(7), 1022–1032. 481. 2000.

46
 International Journal of Computer Applications (0975 – 8887)
Volume 39– No.1, February 2012

[50] Lenard, M. J., & Alam, P. (2004). The use of fuzzy logic detecting fraud in New Zealand, Managerial Auditing
and expert reasoning for knowledge management and Journal 17 (4) 192–204.
discovery of financial reporting fraud. In H. Nemati and [62] Yuan, J., Yuan, C., Deng, Y., & Yuan, C. (2008). The
C. Barko (Eds.), Hershey, PA: Idea Group, Inc. effects of manager compensation and market competition
[51] Lenard, M. J., Watkins, A.L., and Alam, P. (2007). on financial fraud in public companies: an empirical
Effective use of integrated decision making: An study in China, International Journal of Management 25
advanced technology model for evaluating fraud in (2) (2008) 322–335.
service-based computer and technology firms. The [63] Chai, W., Hoogs, B.K., & Verschueren, B.T. (2006).
Journal of Emerging Technologies in Accounting 4(1): Fuzzy Ranking of Financial Statements for Fraud
123-137. detection. In proceeding of International Conference on
[52] Pathak, J., Vidyarthi, N., & Summers, S. L. (2005). A Fuzzy System, (2006), 152–158.
fuzzy-based algorithm for auditors to detect elements of [64] Kiehl, T. R., Hoogs, B. K., & LaComb, C. A. (2005).
fraud in settled insurance claims, Managerial Auditing Evolving Multi-Variate Time-Series Patterns for the
Journal 20 (6) (2005) 632–644. Discrimination of Fraudulent Financial Filings. In Proc.
[53] Bentley, P. (2000). Evolutionary, my dear Watson: of Genetic and Evolutionary Computation Conference,
Investigating Committee-based Evolution of Fuzzy Rules 2005.
for the Detection of Suspicious Insurance Claims. In the [65] Koskivaara, E. (2000). Different pre-processing models
Proc. of GECCO 2000. for financial accounts when using neural networks for
[54] Hoogs, B., Kiehl, T., Lacomb, C., & Senturk, D. (2007). auditing, Proceedings of the 8th European Conference on
A genetic algorithm approach to detecting temporal Information Systems, vol. 1, 2000, pp. 326–3328,
patterns indicative of financial statement fraud, Vienna, Austria.
Intelligent Systems in Accounting, Finance and [66] Feroz, E.H., Kwon, T.M., Pastena, V., & Park, K.J.
Management, 2007, vol. 15: 41-56. (2000). The efficacy of red flags in predicting the SEC's
[55] Juszczak, P., Adams, N.M., Hand, D.J., Whitrow, C., & targets: an artificial neural networks approach,
Weston, D.J. (2008). Off-the-peg and bespoke classifiers International Journal of Intelligent Systems in
for fraud detection‖, Computational Statistics and Data Accounting, Finance, and Management 9 (3) 145–157.
Analysis, vol. 52 (9): 4521-4532. [67] Pacheco, R., Martins, A., Barcia, R.M., & Khator, S.
[56] Bai, B., Yen, J. & Yang. X. (2008). False financial (1996). A hybrid intelligent system applied to financial
statements: characteristics of China's listed companies statement analysis, Proceedings of the 5th IEEE
and CART detecting approach, International Journal of conference on Fuzzy Systems, vol. 2, 1996, pp. 1007–
Information Technology & Decision Making 7 (2) 339– 10128, New Orleans, LA, USA.
359. [68] Liou, F. M. (2008). Fraudulent financial reporting
[57] Deshmukh, A., & Talluru, L. (1998). A rule-based fuzzy detection and business failure prediction models: a
reasoning system for assessing the risk of management comparison. Managerial Auditing Journal Vol. 23 No. 7,
fraud, International Journal of Intelligent Systems in pp. 650-662.
Accounting, Finance & Management 7 (4) 223–241. [69] Kapardis, M. K., Christodoulou, C. & Agathocleous, M.
[58] Deshmukh, A., Romine, J., & Siegel, P.H. (1997). (2010). Neural networks: the panacea in fraud detection?
Measurement and combination of red flags to assess the Managerial Auditing Journal, 25, 659-678.
risk of management fraud: a fuzzy set approach, [70] Ravisankar P., Ravi V., Rao G.R. & Bose I. (2011).
Managerial Finance 23 (6) 35–48. Detection of financial statement fraud and feature
[59] Eining, M. M., Jones, D. R., & Loebbecke, J. K. (1997). selection using data mining techniques. Decision Support
Reliance on decision aids: an examination of auditors‘ System, 50, 491-500.
assessment of management fraud. Auditing: A Journal of [71] Perols, J. (2011). Financial Statement Fraud Detection:
Practice and Theory, 16(2), 1–19. An Analysis of Statistical and Machine Learning
[60] Kotsiantis, S., Koumanakos, E., Tzelepis, D. & Algorithms, Auditing: A Journal of Practice and Theory,
Tampakas, V. (2006). Forecasting fraudulent financial Vol. 30(2), pp. 19-50.
statements using data mining, International Journal of [72] Zhou, W. and Kapoor, G. (2011). Detecting evolutionary
Computational Intelligence 3 (2) 104–110. financial statement fraud. Decision Support Systems, v.
[61] Owusu-Ansah, S., Moyes, G.D. , Oyelere, P.B., Hay, P. 50, n. 3, p. 570-575.
(2002). An empirical analysis of the likelihood of

47