# Tax Fraud Reduction Using Analytics in an East European Country

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Conceptual Model for Tax Obligations’ Risk Management

#### 2.1. Taxpayer Grouping

- the type of economic activity;
- payable fees and their amount (e.g., income tax, VAT, excise duties, and others);
- taxable person specifics (e.g., small businesses, social enterprises, companies established in free economic zones, and others, which are subject to different tax rates and/or taxation rules for corporate tax);
- other characteristics or combinations thereof.

- Taxpayers who honestly perform tax obligations;
- Taxpayers who strive to honour their tax obligations fairly but not always successfully;
- Taxpayers are generally inclined not to comply with their tax obligations until they are addressed;
- Taxpayers who maliciously avoid fulfilling tax obligations.

#### 2.2. Taxpayer Grouping

- registration risk (the person does not register as a taxpayer in time, does not deregister from the taxpayer register, or registers for other than economic purposes);
- the risk of incorrect tax declaration;
- the risk of delayed tax declaration when filing tax returns is delayed, or declarations are not provided at all;
- tax payment risk, where the payment of taxes is delayed or the fees are not paid.

- avoidance of registration of ongoing economic activity;
- avoidance of taxpayer registration and registration for other than economic activities;
- avoidance of declarations of income and taxes;
- incorrect application of illegal use of tax relief;
- avoidance of labour-related taxes;
- delay in declaring taxes;
- non-payment of taxes.

#### 2.3. Tax Risk Analysis Model

- Square 1—Strategic Focus Level

- Square 2—Aggregated Focus

- Square 3—Case Focus

#### 2.4. Evaluation of Key Risks at the Aggregated Analysis Stage

#### 2.5. Measures to Manage the Risk of Tax Noncompliance

- Taking into account the indicators for assessing the efficiency of taxpayer control processes and evaluating the effectiveness of control or monitoring actions in specific sections (e.g., the execution time of control or monitoring actions, number of irregularities detected, amount of charges charged, amount of fees received, the change in indicators for the sector, where the actions were applied, and others);
- Comparison of actual results of control or monitoring actions with expected results (e.g., comparing expected results with actual results at the end of the measure upon completion of control or monitoring);
- Comparing the procedures for completing specific control actions established by law (for example, a tax investigation ends up with a recommendation for the taxpayer to pay an additional fee and to verify the declarations, while the tax verification is already completed in an act where the additional charge is calculated and recorded as an obligation after subsequent procedures), the duration of actions (established by legislation, actual average, and others) as well as the results of the application of the different tendency to offend against taxpayer segments;
- Identify and track indicators to assess the effectiveness of taxpayer control and monitoring actions, such as key performance indicators;
- Depending on the possible ways of evaluating the effectiveness of control or monitoring actions, the control or monitoring activities results can be compared with the control group at different periods, and different segments and differences from target values can be assessed. It should be noted that the effectiveness of taxpayers’ controls or monitoring activities can only be assessed by providing high-quality, comprehensive, and accessible information about the control and monitoring tools applied and the results acquired. The selection of optimal taxpayer control or monitoring tools and the scope and duration determination is only possible with the above-denoted information and complex data analysis.

## 3. Software and Its Functionality

^{®}Intelligence Platform, which consists of the following levels: databases, activity logic, and user (see Figure 7).

^{®}language, SAS

^{®}Enterprise Guide

^{®}, SAS Data

^{®}Integration Studio) and logical levels (SAS

^{®}Information Map Studio, SAS

^{®}Financial Management Studio). All the denoted technologies provide the ability to create the necessary data models and software solutions independently.

^{®}Metadata Server, consists of three layers:

- Database layer—SAS
^{®}Metadata Server Database; - User layer—SAS
^{®}Management Console. SAS^{®}Management Console is the system end-user tool for viewing and modifying the activity logic model using standard activity terminology; - SAS
^{®}Metadata Model for Activity Logic. SAS^{®}Metadata Model is a hierarchical object model defining metadata types and their associations. The SAS^{®}Metadata Model links the database layer to the user layer; - Software performance analysis component consists of;
- Forecasting module—SAS
^{®}Financial Management; - Free-form analysis module—SAS
^{®}Enterprise Guide^{®}; - View module—SAS
^{®}Visual Analytics; - Standard report generation and distribution module—SAS
^{®}Web Report Studio; - Key performance indicators and their hierarchies module—SAS
^{®}Information Delivery Portal and SAS^{®}Web Report Studio; - Microsoft Office integration module—SAS
^{®}Add-In for Microsoft Office.

^{®}Enterprise Miner™ and SAS

^{®}Visual Analytics. These products have different uses and are intended for different user groups. SAS

^{®}Enterprise Miner™ is designed for advanced analysts and users with strong mathematical–analytical skills. SAS

^{®}Visual Analytics is designed for data analysis, “friendly” analytics, and reporting. SAS

^{®}Visual Analytics is designed for a wider range of users and does not require deep statistical knowledge from users. SAS

^{®}Enterprise Miner™ is based on the SEMMA (Sample, Explore, Modify, and Assess) methodology that allows for performing primary and statistical data mining, data preparation, and model accuracy estimation actions.

## 4. Adapting Analytical Methods to Risk Management of Tax Obligations

#### 4.1. Data Grouping and Clustering Methods

#### 4.1.1. Data Grouping

- a study is being carried out to find out the different characteristics and needs of taxpayers;
- taxpayers are combined according to their characteristics and needs (the process is complex because the same environment can be grouped in various ways, often leading to the discovery of new taxpayer groups. This step ensures a more accurate assessment of taxpayers’ wishes and provides targeted information);
- the most appropriate (target) groups are selected, and irrelevant or inappropriate groups for the tax administrator’s strategic goals are rejected.

#### 4.1.2. Hierarchical Clustering

#### 4.1.3. Center-Based Clustering

- Cluster centroids initialization: randomly choose k different points as initialized centroids ${C}_{1},{C}_{2}\dots {C}_{k}$ for $k$ groups, where ${C}_{k}$ is d-dimension vector $\left({C}_{k1},{C}_{k2}\dots {C}_{kd}\right),k\in \left[K\right]$;
- Repeat the following until the stopping criterion:
- (a)
- For $i\in \left[n\right],k\in \left[K\right]$, compute the Euclidean distance between point ${x}_{i}$ and centroids ${C}_{k}$ by$${X}_{ik}=\sqrt{{\displaystyle \sum}_{j=1}^{d}{\left({x}_{ij}-{C}_{kj}\right)}^{2}}.$$
- (b)
- Assign each data point ${x}_{i}$ to the closest cluster ${m}_{i}$ for $i\in \left[n\right]$. This can be carried out by computing ${k}_{i}\leftarrow \mathrm{arg}min\left\{{X}_{i1},{X}_{i2},\cdots ,{X}_{ik}\right\}$ firstly, and then generate a k-dimension one-hot vector ${b}_{i}$ where ‘1′ indicates the ${k}_{i}$-th component of vector $\left({X}_{i1},{X}_{i2},\cdots ,{X}_{ik}\right)$. We form $K\times n$ matrix $B$ such that the i-th column of $B$ is the one-hot vector ${b}_{i}$. Let ${m}_{k}$ be the k-th row of $B$.
- (c)
- Recalculate the average of the points in each cluster. For each cluster $k\in \left[K\right]$, compute new cluster center with$${\phi}_{k}=\frac{{m}_{k}\xb7x}{{\chi}_{k}},$$
- (d)
- Check the stopping criterion and update the new cluster center with the average. For each $k\in \left[K\right]$, compute the Euclidean distance between φ
_{k}and ${C}_{k}$ at first, and then the squared error can be computed by$$e={\displaystyle \sum}_{k=1}^{K}{e}_{k}={\displaystyle \sum}_{k=1}^{K}\sqrt{{\displaystyle \sum}_{j=1}^{d}{\left({\phi}_{kj}-{C}_{kj}\right)}^{2}}.$$

#### 4.1.4. Spherical K-Means Clustering

- ${d}_{ic}^{\u2033}=1$ if observation $i$ is in cluster $c$;
- ${d}_{ic}^{\u2033}=0$ otherwise.

- ${d}_{ih}^{\prime}=\frac{1}{{n}_{c}}$ if observations $i$ and $h$ are in cluster c;
- ${d}_{ih}^{\prime}=0$ otherwise.

#### 4.2. Classification Methods

#### 4.2.1. Decision Trees

- Identify objects (e.g., taxpayers) according to their potential dependence on a particular classification group;
- Assign objects (e.g., taxpayers) to a specific category, such as low, medium, and high-risk groups;
- Predict future events, such as tax evasion, based on the model that was developed;
- Compress a large group of available data, leaving independent variables with only a statistically significant impact on the dependent variable prediction;
- Identify interactions between individual test groups.

#### 4.2.2. Logistic Regression

- Binary logistic regression. The target variable is binary. For example, while investigating what determines the attitude towards tax evasion (need–no need), we want to find out what the decision to fake financial accounts depends on;
- Multiple logistic regression. The target variable is categorical, but it acquires more than two values. We want to find out what influences the choice of the pension insurance fund and what determines the voting priorities (which party to choose);
- Ordinal logistic regression. The target variable is ordinal, with values indicating an increasing (decreasing) amount of some property. For example, to decide what determines the evasion of some tax, all taxpayers are divided into groups who avoided the tax during the first year after the start of business, during the second year, and more than two years later.

#### 4.2.3. Support Vector Machine

- Identifying objects (such as taxpayers, invoices, and others) according to their potential dependence on a particular classification group;
- Assigning items (such as taxpayers, and invoices) to a specific category, such as low, medium, and high-risk groups;
- Predicting future events, such as tax evasion, according to the created model.

- First, it has a control parameter that avoids overfitting;
- Second, it uses the kernel property, so it can be built using problem-related expert knowledge;
- Third, the support vector machine method, as with other effective methods, defines the problem of convex optimization (not a local minimum);
- Finally, SVM approximates the level of test error within the boundary.

#### 4.2.4. Neural Networks

- $x={\left(1,{x}_{1},{x}_{2},\dots ,{x}_{r}\right)}^{\prime}$ are the network inputs (independent variables), where 1 corresponds to the bias of a traditional model;
- ${\gamma}_{j}={\left({\gamma}_{j0},{\gamma}_{j1},\dots ,{\gamma}_{ji},\dots ,{\gamma}_{jr}\right)}^{\prime}\u03f5{\Re}^{r+1}$ are the weights of the inputs layer neurons to those of the intermediate or hidden layer;
- ${\beta}_{j},j=0,\dots ,q,$ represents the hidden units’ connection force to those pertaining to output ($j=0$ indexes the bias unit), and $q$ is the number of intermediate units, that is, the number of hidden layer nodes;
- $W$ is a vector that includes all the synaptic weights of the network, ${\gamma}_{j}$ and ${\beta}_{j}$, or connections pattern;
- $Y=\widehat{f}\left(x,W\right)$ is the network output;
- $F:\Re \to \Re $ is the unit activation function and output while $G:\Re \to \Re $ corresponds to the intermediate neurons activation function.

#### 4.3. Methods Based on Dependencies and Logical Relations

#### 4.3.1. Multidimensional Regression

- Defining interrelations between different indicators, for example, by examining taxpayers’ behavioural habits;
- Predicting future events, such as tax evasion.

#### 4.3.2. Association Rules

#### 4.4. Other Data Mining Methods

#### 4.4.1. The Selection of Important Features

- Latent features do not always exist and can not always be reliably distinguished;
- Different sets of feature analysis applied to the same data produce different sets of possible features;
- Excluded features are not always easy to interpret.

#### 4.4.2. Anomaly Detection

- Model outliers are the dependent variable values that are “sharply” different from the values predicted by the model. This deviation is usually measured by standardized model errors, which can be defined differently. One way would be to rely on the values of the diagonal elements of the hat matrix;
- Observations that are “far away” from the “centre” of predictive variables (i.e., have a high leverage) may have a potentially greater influence on regression coefficients. One of the most popular characteristics of observation influence is its hat value;
- High-leverage observations, which also have the feature of exclusivity, are influential because, by removing them from the model, the coefficients significantly change. Cook’s distance is often used to measure the change.

#### 4.5. The Application of Clustering and Classification Methods in the Field of Taxpayer Control and Monitoring

#### 4.6. Model Selection Criteria

## 5. Experimental Results

#### 5.1. Segmentation Model Based on Taxpayers’ Behaviour

#### 5.1.1. Variable Selection

- From a business perspective, i.e., according to the intended use of the model, the selected variables must be meaningful;
- Variables influence the structure of the segments (verified by preliminary analysis of the created model);
- Time-stable variables.

- If the value of the significance level corresponding to the Kolmogorov statistic is less than 0.05, then the indicator is rejected;
- If the significance level of a Kolmogorov statistic corresponding to the indicator is between 0.05 and 0.1, the indicator can be included or rejected—the decision must be made according to other criteria;
- If the significance level of Kolmogorov’s statistic is more than 0.05, the indicator must be added to the modelling.

#### 5.1.2. Variable Transformations

#### 5.1.3. Modelling Results

- About 20 significant variables have to be selected for modelling (in the case of the sample presented in this study, the main components were used);
- Segment structure should consist of 4–7 segments;
- The segments must be homogeneous according to the selected attributes and differ from each other;
- For better understanding and recognition, segments should be ‘measurable’ in size and identified by their characteristics;
- It must be possible to write the characteristics of each segment in one sentence.

#### 5.2. VAT Assessment Model for VAT Payer Checkout

- Decrease in share capital;
- Legal status change (e.g., liquidation, bankruptcy, and reorganization);
- Change in the head of the company;
- Significant change (decrease) in the number of employees;
- Significant decrease in revenue;
- Significant decrease in assets;
- Etc.

#### 5.2.1. Analytical Data Flow Process

**Sampling**: A representative sample selection method, namely separate sampling was used to solve the prediction task. The sample included 808 deregistered and 1616 registered taxpayers.

**Data partitioning**: To create the most appropriate model, the sample is randomly (by applied stratification) partitioned into training and validation subsets. A total of 564 deregistered and 1130 registered taxpayers were enrolled for training, and, respectively, 244 and 486 taxpayers for validation.

**Selection of variables**: according to the model’s purpose, this step’s objective is to select significant variables. In general, they should meet the following requirements:

- be selected from a business point of view, i.e., significant variables according to the intended use of the model;
- variables should have a significant impact on the prediction result;
- be time-stable variables.

#### 5.2.2. Significance and Efficiency of the Model

## 6. Conclusions and Future Works

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Conflicts of Interest

## References

- Allingham, M.G.; Sandmo, A. Income tax evasion: A theoretical analysis. J. Public Econ.
**1972**, 1, 323–338. [Google Scholar] [CrossRef] [Green Version] - Murphy, R. Closing the European Tax Gap; A Report for Group of the Progressive Alliance of Socialists & Democrats in the European Parliament; Tax Reasearch LLP: Ely, UK, 2012. [Google Scholar]
- Siti, N.O.S.; Bojuwon, M. Reengineering Tax Service Quality Using a Second Order Confirmatory Factor Analysis for Self-Employed Taxpayers. Int. J. Trade Econ. Financ.
**2014**, 5, 429–434. [Google Scholar] - Veermäe, E. Improving Audit Efficiency: Estonian Tax and Customs Board Uses SAS to Detect and Prevent Tax Fraud; SAS Institute White Papers; SAS: Cary, NC, USA, 2015. [Google Scholar]
- Fiscalis Risk Analysis Project Group. Risk Management Guide. For Tax Administrations (FRAPG); European Commision: Brussels, Belgium, 2016. [Google Scholar]
- Bonchi, F.; Gianotti, F.; Mainetto, G.; Pedreschi, D. A Classification-Based Methodology for Planning Audit Strategies in Fraud Detection. In Proceedings of the Fifth ACM SIGKDD International Conference, San Diego, CA, USA, 15–18 August 1999; pp. 175–184. [Google Scholar]
- Hinton, G.; Deng, L.; Yu, D.; Dahl, G.E.; Mohamed, A.; Jaitly, N.; Senior, A.; Vanhoucke, V.; Nguyen, P.; Sainath, T.N.; et al. Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups. IEEE Signal Process. Mag.
**2012**, 29, 82–97. [Google Scholar] [CrossRef] - Sabău, A.-I.; Mare, C.; Safta, I.L. A Statistical Model of Fraud Risk in Financial Statements. Case for Romania Companies. Risks
**2021**, 9, 116. [Google Scholar] [CrossRef] - Schneider, F.G. Tax Losses Due to Shadow Economy Activities in OECD Countries from 2011 to 2013: A Preliminary Calculation; CESifo Working Paper Series No. 5649; CESifo: Munich, Germany, 2015. [Google Scholar]
- Woodward, A.; Vesset, D. Worldwide Advanced and Predictive Analytics Software Market Shares, 2014: The Rise of the Long Tail; IDC: Needham, MA, USA, 2014; pp. 1–7. [Google Scholar]
- Kuan, C.M.; White, H. Artificial neural networks: An econometric perspective. Econom. Rev.
**1994**, 13, 1–91. [Google Scholar] [CrossRef] - Maasoumi, E.; Khotanzad, A.; Abaye, A. Artificial neural networks for some macroeconomic series: A first report. Econom. Rev.
**1994**, 13, 105–122. [Google Scholar] [CrossRef] - Nelson, C.R.; Plosser, C.I. Trends and random walks in macroeconmic time series: Some evidence and implications. J. Monet. Econ.
**1982**, 10, 139–162. [Google Scholar] [CrossRef] - Campbell, J.Y.; Lo, A.W.; MacKinlay, A.C. The Econometrics of Financial Markets; Princeton University Press: Princeton, NJ, USA, 1996. [Google Scholar]
- Cover, J.P. Asymmetric Effects of Positive and Negative Money Supply Shocks. Q. J. Econ.
**1992**, 107, 1261–1282. [Google Scholar] [CrossRef] - Hutchinson, J.M.; Lo, A.W.; Poggio, T. A Nonparametric Approach to Pricing and Hedging Derivative Securities via Learning Networks. J. Financ.
**1994**, 49, 851–889. [Google Scholar] [CrossRef] - Morgan, D.P. Asymmetric Effects of Monetary Policy. Fed. Reserve Bank Kans. City Econ. Rev.
**1993**, 78, 21–33. [Google Scholar] - Muller, P. The Information Content of Financial Aggregates During the 1980s. In Monetary Seminar 90; Bank of Canada: Ottawa, ON, Canada, 1990; pp. 183–304. [Google Scholar]
- Rossomando, F.G.; CM Soria, C.M. Adaptive neural sliding mode control in discrete time for a SCARA robot arm. IEEE Lat. Am. Trans
**2016**, 14, 2556–2564. [Google Scholar] [CrossRef] - Torgler, B. Tax Morale and Compliance Review of Evidence and Case Studies for Europe; The World Bank, Europe and Central Asia Region Human Development Economics Unit: Washington, DC, USA, 2011. [Google Scholar]
- Parker, D.B. Learning Logic: Technical Report TR-47; Center for Computational Research in Economics and Management Science, MIT: Cambridge, MA, USA, 1985. [Google Scholar]
- Raub, B.G.; Chen, W.W. A Cluster Analysis Approach To Describing Tax Data. Stat. Income SOI Bull. Res. Pap.
**2005**, 25, 158–159. [Google Scholar] - Zhao, L.; Lee, S.; Jeong, S.-P. Decision Tree Application to Classification Problems with Boosting Algorithm. Electronics
**2021**, 10, 1903. [Google Scholar] [CrossRef] - Gonzales, P.C.; Velasquez, J.D. Characterization and detection of taxpayers with false invoices using data mining techniques. Expert Syst. Appl.
**2013**, 40, 1427–1436. [Google Scholar] - Pérez López, C.; Delgado Rodríguez, M.J.; de Lucas Santos, S. Tax Fraud Detection through Neural Networks: An Application Using a Sample of Personal Income Taxpayers. Future Internet
**2019**, 11, 86. [Google Scholar] [CrossRef] [Green Version] - Serrano, A.M.R.; da Costa, J.P.C.L.; Cardonha, C.H.; Fernandes, A.A.; de Sousa Júnior, R.T. Neural Network Predictor for Fraud Detection: A Study Case for the Federal Patrimony Department. In Proceedings of the International Conference on Forensic Computer Science (IJoFCS), Brasilia, Brazil, 26–28 September 2012; pp. 61–66. [Google Scholar]
- Xu, Z. Logistic Regression Based on Individual-Level Predictors and Aggregate-Level Responses. Mathematics
**2023**, 11, 746. [Google Scholar] [CrossRef] - Challenges and Implementation Measures of the State Tax Inspectorate (CIMSTI); State Tax Inspectorate Under the Ministry of Finance of the Republic of Lithuania: Vilnius, Lithuania, 2014.
- Smedegård, O.Ø.; Jonsson, T.; Aas, B.; Stene, J.; Georges, L.; Carlucci, S. The Implementation of Multiple Linear Regression for Swimming Pool Facilities: Case Study at Jøa, Norway. Energies
**2021**, 14, 4825. [Google Scholar] [CrossRef] - Managing and Improving Compliance: Recent Developments in Compliance Risk Treatments (MIC); OECD: Paris, France, 2009.
- Monitoring Taxpayer’s Compliance: A Practical Guide Based on Revenue Body Experience (MTC); OECD: Paris, France, 2008.
- Compliance Risk Management Guide for Tax Administrations (CRMGTA); Fiscalis Risk Management Platform Group, European Union: Brussels, Belgium, 2010.
- Rhee, W.; Rich, R.W. Inflation and the Asymmetric Effects of Money on Output Fluctuations. J. Macroecon.
**1995**, 17, 683–702. [Google Scholar] [CrossRef] - Liu, X.; Pan, D.; Chen, S. Application of Hierarchical Clustering in Tax Inspection Case-Selecting. In Proceedings of the 2010 International Conference on Computational Intelligence and Software Engineering (CiSE), Wuhan, China, 10–12 December 2010; pp. 1–4. [Google Scholar]
- Lückeheide, S.; Velásquez, J.D.; Cerda, L. Segmentacion de los Contribuyentes que Declaran IVA Aplicando Herramientas de Clustering. Rev. Ing. Sist.
**2007**, 21, 87–110. [Google Scholar] - Poco, M.; Lopes, C.; Silva, A. Perception of Tax Evasion and Tax Fraud in Portugal: A Socialogical Study; OBEGEF—Observatório de Economia e Gestão de Fraude (OBEGEF Working Papers on Fraud and Corruption): Porto, Portugal, 2015. [Google Scholar]
- Li, Z. Research on Distinguish the Accounting Information Distortion Based on the Principle Components-logistic Regression Model. Int. J. Secur. Its Appl.
**2014**, 8, 37–50. [Google Scholar] [CrossRef] - Devos, K. The Influence of Tax Preparers upon Individual Taxpayer Compliance-Some Australian Evidence. In Proceedings of the 24th Australasian Tax Teachers Association Conference, Sydney, Australia, 16–18 January 2012; p. 38. [Google Scholar]
- DeBarr, D.; Wechsler, H. Fraud Detection Using Reputation Features, SVMs, and Random Forests. In Proceedings of the 9th International Conference on Data Mining, Las Vegas, NV, USA, 22–25 July 2013; pp. 238–244. [Google Scholar]
- Shams, R.; Mercer, R.E. Supervised classification of spam emails with natural language stylometry. Neural Comput. Appl.
**2016**, 27, 2315–2331. [Google Scholar] [CrossRef] - Poh, H.L.; Yao, J.; Jašic, T. Neural Networks for the Analysis and Forecasting of Advertising and Promotion Impact. Intell. Syst. Account. Financ. Manag.
**1998**, 7, 253–268. [Google Scholar] [CrossRef] - Alstadsæter, A.; Jacob, M. The Effect of Awareness and Incentives on Tax Evasion; CESifo Working Paper Series No. 4369; CESifo: Munich, Germany, 2013; p. 42. [Google Scholar]
- State Tax Inspectorate Under the Ministry of Finance of the Republic of Lithuania (SIT). Available online: https://www.vmi.lt/index.jsp?lang=en (accessed on 25 February 2023).
- Moeinadin, M.; Heirany, F.; Harandi, A. Identifying the Effective Factors for the Improvement of Tax Compliance. Int. J. Acad. Res. Account. Financ. Manag. Sci.
**2014**, 4, 220–227. [Google Scholar] [CrossRef] [Green Version]

**Figure 4.**The execution of tax obligations [28].

SEGMENT ▲ | 1 | 2 | 3 | 4 | 5 |
---|---|---|---|---|---|

1 | 0 | 8.010 | 6.687 | 5.200 | 7.345 |

2 | 8.010 | 0 | 4.845 | 5.141 | 11.841 |

3 | 6.687 | 4.845 | 0 | 4.127 | 10.613 |

4 | 5.200 | 5.141 | 4.127 | 0 | 9.880 |

5 | 7.345 | 11.841 | 10.613 | 9.880 | 0 |

Fit Statistics | Train | Validation |
---|---|---|

AIC | 1810.56 | |

ASE | 0.176804 | 0.190389 |

SBC | 1870.34 | |

MSE | 0.177858 | 0.190389 |

RMSE | 0.421733 | 0.436336 |

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |

© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Ruzgas, T.; Kižauskienė, L.; Lukauskas, M.; Sinkevičius, E.; Frolovaitė, M.; Arnastauskaitė, J.
Tax Fraud Reduction Using Analytics in an East European Country. *Axioms* **2023**, *12*, 288.
https://doi.org/10.3390/axioms12030288

**AMA Style**

Ruzgas T, Kižauskienė L, Lukauskas M, Sinkevičius E, Frolovaitė M, Arnastauskaitė J.
Tax Fraud Reduction Using Analytics in an East European Country. *Axioms*. 2023; 12(3):288.
https://doi.org/10.3390/axioms12030288

**Chicago/Turabian Style**

Ruzgas, Tomas, Laura Kižauskienė, Mantas Lukauskas, Egidijus Sinkevičius, Melita Frolovaitė, and Jurgita Arnastauskaitė.
2023. "Tax Fraud Reduction Using Analytics in an East European Country" *Axioms* 12, no. 3: 288.
https://doi.org/10.3390/axioms12030288