Machine Learning in Auditing: Problems, Solutions, Guidelines,

Future Directions

Xinke Bai

International Institute of Joint Audit, Nanjing Audit University, Nanjing, China

Keywords: Auditing, Machine Learning (ML), Anomaly Detection, Financial Transparency, Data Privacy.

Abstract: In the modern and extremely dynamic business arena, auditing can be seen as the only way of getting financial

information clarified, taking individuals who have that responsibility to account, and creating the impression

of an honest player. The effectiveness of auditing is not just limited to risk mitigation approaches considering

prevention of fraud, as it also strengthens the principles of legal compliance, market order, and economic

stability in general. Several factors are propelling the auditing industry through a significant transformation

at an increasingly fast rate, and these factors are big data and AI, of which machine learning (ML) is one of

them. Traditional audits are made faster and more precision by ML through scrutinizing and interpreting

enormous and complex data sets, such as detecting risk and anomaly violations and automatically generating

reports. Yet, the auditing activities introduce ML into the auditing processes this creates another problem.

These encompass issues of data quality and reliability, models transparency and interpretability for complex

systems, risks of overfitting and underfitting, and also include security and moral considerations in the

software and hardware regulations. On the other hand, there exists a cognate challenge of the auditors getting

upskilled to effectively exploit these AI-driven technologies. This paper seeks to determine the problems of

empirical evaluation of ML techniques and suggest possible solutions to maximize the efficiency and

productivity of the auditing processes.

1 INTRODUCTION

Audit is one of the keystones of the present-day ever

increasingly complicated and variable business

environment, and audit should be seen as the

foundation for pointing out the gaps in financial

transparency and reliability.

Although it is the assessment of a company’s

performance measure, auditing is also important for

fraud detection and prevention, internal control

improvement, and risk management. The

Implications of necessary auditing are apparent as a

means of legal compliance, maintaining the market

order, and the growth of the economy. In an age

where information can be tantamount to money,

auditing helps to maintain a public trust and assure

the market standard.

The very apace at which technology is now

manifesting itself, especially in data collection and

artificial intelligence, has made machine learning

(ML) emerge as a proverbial tool in a number of

https://orcid.org/0009-0009-8547-1349

industries. From self-driving cars to product

recommendation systems, ML technology has

evolved so much that it can analyze and correlate data,

which is fundamentally redefining the conventional

practices. In the case of the auditing sector, which

depends heavily on data correctness and speed, ML

can save so much effort. Inside the audit process, the

application of such techniques as anomaly detection,

risk assessment, and automated report generation can

help boost the efficiency and precision of audit

outcomes significantly. A perfect example is ML; this

technique can monitor financial transactions

continuously in real time to detect any variations,

provide predictive risk analysis, and generate a

complete report about the audit, thus, saving human

errors and increasing productivity (Hassan, 2023).

However, the implementation of Machine

Learning in audits may not be without its difficulties,

such as the ones associated with data quality, model

transparency, overfitting and shifting characteristics,

regulatory compliance, or the auditors’ skills. A field

496

Bai, X.

Machine Learning in Auditing: Problems, Solutions, Guidelines, Future Directions.

DOI: 10.5220/0013269400004568

In Proceedings of the 1st International Conference on E-commerce and Artiﬁcial Intelligence (ECAI 2024), pages 496-503

ISBN: 978-989-758-726-9

of auditing that focuses on the implementation of

machine learning identifies five issues: data quality

and visualization, model interpretability and

transparency, overfitting and generalization,

regulatory compliance, and ethics, the skills and

training gap among auditors. The section proposes the

measures to improve data governance, model

interpretability, overfitting control, compliance and

ethical concerns, and auditor's technical skills all of

which would contribute to the solution of these issues.

Such methods are likely to contribute to the

subsequent diligence of machine learning in auditing,

resulting eventually in the growth of the quality and

efficiency of audits. This article tries to deliver

valuable findings and to present some

recommendations of efficiently using ML for raising

the level of audits’ quality and providing compliance

with the laws of compliance.

2 CHALLENGES

2.1 Data Accuracy and Precision

Another major feature concerning the auditors is the

incomplete data that they get. The summary part of

the report may be distorted in most cases when the

transaction data, such as date, time, nature of

transaction, cost of goods sold, etc., are revealed

incompletely or inadequately. These areas inhibit

machine learning from discerning anomalies, fraud

and compliance violations; hence, the level of audit

clutter cannot be trusted and lacks reassurance (Office

of Inspector General, 2012). The utilization of

machine learning on unsuitable and incomplete data

sets will not only yield inaccurate insights, but also

miss out on knowing the profound risks and

anomalies the artificial intelligence should have been

dealing with.

Discrepant data is one of the major difficulties

confronting practitioners. Disparities in data entry

systems, human mistake, and variable specificity of

data sets could be the cause of drifted information

between data source systems. Such differences can

disrupt the engine of machine learning algorithms that

form the basis of this whole idea, thereby provoking

errant conclusions about risk assessment or wrong

identification of mischief during the audit (Stefanov

et al., 2012). For instance, labelling the same financial

transactions originated from one source into different

categories could mean that the machine will not be

able to notice the patterns of fraud or error.

Additionally, machine learning is further

complicated because of wrong data, like

discrepancies in the transaction amounts or

accounting transactions that are not in the right

account. Such errors will result in a situation where

the models are used to learn from the wrong

information, and their outputs may not be useful

inputs to even monetizing the said organization.

Maintaining high quality and integrity of data

becomes increasingly challenging but critical,

particularly with businesses generating huge volumes

of complex data.

In an environment where auditing becomes too

demanding for the teams due to these data quality

matters, the entire audit process may fail leading to

serious mistakes in audit conclusions, and, moreover,

it can result to failing to meet the compliance and

regulatory requirements. Hence, the pillar that goes

hand in hand with the capacity to utilize machine

learning so as to improve audit accuracy, efficiency,

and effectiveness is data quality and integrity

(Stefanov, 2014). If these problems are not solved,

many of the advantages of ML-driven practices will

remain unexploited, and the overall audit result

quality will be compromised.

2.2 ML Interpretation and Describe

An enormous difficulty to use machine learning in the

auditing field is the issue of machine learning models

transparency and interpretability. The advanced

machine learning models, especially deep learning

that are the most robust algorithms nowadays, tend to

be “black boxes” thus having no knowledge of the

way the decision-making is based on (Carvalho,

Pereira, & Cardoso, 2019). These models utilize data

processing as well as decision making systems that

enable the algorithms to successfully do that, but the

humans, including the auditors, cannot follow the

internal procedures used for arriving at the decisions

because such procedures are not clearly shown to

them.

In auditing, this kind of uncertainly may be

problematic since the auditors need to know the path

of how decisions have been taken to achieve every

conclusion. In finance, traditional auditing entails the

following and logically verifying procedures and

steps that are easy to understand. However, the

process of machine learning has been found to be

complex, and it is not easy to follow the road from

input data to output decision, especially when there

are deep neural networks with numerous layers in

between. This complexity impedes auditors from

interpreting the reasoning behind the model outputs

to validate and attest audit findings which is one of

the most important steps in the whole audit process.

Machine Learning in Auditing: Problems, Solutions, Guidelines, Future Directions

497

Algorithmic systems’ “black box” nature is usually

taxing for an auditor because it becomes impossible

to correctly illustrate the basis for unearthing the

fraud or signs of suspicion. Having no specific

understanding of how a model generates results

presents an issue for auditors since this will make it

quite hard to ensure the reliability and are correct

these results (Knechel, 2013). This lack of clarity may

weaken supposed confidence and truthful audit trails

of the audit process, which the stakeholders rely on

for the credibility of this profession.

2.3 Overfitting and Generalization

Overfitting is notoriously the biggest hazard that a

machine learning model is in possession of,

especially if this model is used to provide a complex

task like auditing. In a nutshell, one type of the

needless compliance with the training data appears to

happen when the model captures the data set which it

is already familiar with, namely capturing the general

patterns beside the noises and anomalies present in

that data set (Adler et al., 2018). Now, the model,

which was otherwise perplexed by the training data,

does almost perfectly on this new data set. So, it can't

learn or maybe draw inferences from this data set

which is not seen yet. The paraphrase of this is that

the production of genuine data would be

compromised by this situation of overfitting, and

subsequently cannot be used in real time, which

involves not only diverse but also varying data among

different audit engagements.

In shapes of self-regulatory conflicts, overfitting

to be found is more particular in the case of audit

inspection. Auditors lag on ML algorithms and

machines for identifying the fraudulent activities,

estimating risk, and coming up with decisions based

on historical and current financial data. Nonetheless,

an overzealous working model that fails to see the

randomness or specific pattern variants that are

duplicates of the training data set can lead to false

positives or false negatives. E.g., it might be found to

incorrectly flag ordinary transactions to be fraudulent

if they have the resemblance to the distortions of the

training data, or miss out the real fraudulent

transactions, which might appear to have the slight

dissimilarities from the patterns the ML has learned.

Generalizing is made even more difficult by the

fact that there are many different types of financial

data. Undertaking of auditing purposely means

dealing with diverse data from different clients,

industry, or time, which all have differing

characteristics. Striking a right balance between a

model that is hyper-focused on one set of data and

widely enveloping datasets can make the audit

process reliable across various audit scenarios (Adler

et al., 2018). This heterogeneity renders the machine

learning models apart from being pre-programmed to

learn from past information, they need to be pre-

programmed to move freely within the real world and

handle multiple and diverse situations.

2.4 Regulating Coherently and

Ethically

Machine learning in the audit process introduces

substantial challenges with respect to regulatory

compliance, data privacy, security, and ethical

dimensions. The advancement of the machine

learning models, which are increasingly applied in the

analysis of sensitive financial data and in making

decisions that influence many stakeholders, imposes

strict standards of ethics so that the regulation

provides transparency, fairness, and customer

satisfaction.

Another important aspect is the data privacy that

is often prioritized and involve large volume of

sensitive information with the clients, employees, and

customers that is recorded on paper (Criado, Ferrer,

& Such, 2021). The utilization of machine learning

also becomes dependent on the ability to access and

process the sensitive data; therefore, the sensitive data

would be at risk of accidental data breaches and

unauthorized access. People with data privacy in

mind may not be comfortable with complying with

the strict rules such as the GDPR in the EU of their

personal data (Fasterling 2012). Machine Learning

algorithms should thus employ techniques like

pseudonymization, encryption, and additionally

should have restricted access to the data by giving

only the authorized personnel right of use.

Security is a fundamental element of strategy.

Training models of machine learning demand the

robust datasets that must be kept in a safe

environment. It is within these datasets that probing

and hacking may plague the system and thereby result

to the undermining of the model outputs and therefore

the audit results could be incorrect. Auditing entities

are liable to build up resilient cybersecurity measures

as the chief focus of thwarting any attacks by an

organization and to guard their data infrastructure in

a safe manner.

2.5 Education of Auditors and Skills

Riding on the border of progress in indexing machine

learning into auditing is the gap in skills between the

auditors. Historically, auditors defined themselves as

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence

498

financial analytics, risk assessors, and compliance

auditors. However, there has been a simulation of AI

in the current auditing system, and this has led to the

addition of the technical form skills. These skills are

understanding how machine learning models involve,

interpreting their outcome, and detecting issues like

encoding biases and model overfitting (Fasterling,

2012). Currently, many auditors are not adequately

knowledgeable about these advanced tools, which can

be a big deterrent to leveraging the utmost of these

technologies, thus hampering the overall audit

operation.

The speed at which new innovations in machine

learning technologies are coming in increases the

skills gap of those being trained and currently being

tolerated for auditors. In conclusion, AI tools for

measurement can create a lot of problems for an

auditor who is not ready to meet a need for complex

tasks include such as decision-making, risk

management, and control. This situation brings about

so many issues. Among the many other possibilities

is the auditors not inspecting the machine learning

models’ outputs well and the assumptions and

limitations inherent in the models; instead; they

would take the model’s outputs for granted and this

might end with false audit conclusions.

The problems also involve the auditors who may

have insufficient or deficient training; hence, they

may have difficulties in interpreting the machine

learning models’ outputs. By contrast, the accounting

principles that categorically specific each traditional

audit procedures are not used in machine learning

models. That is because it runs as a “black boxes,”

looking for the patterns in data that managers or

auditors may not directly know (Cao, 2017). Non-

specialized auditors in machine learning may have

difficulties in forming audit findings that could be

easily understandable, which is a counterbalance for

why audit can be traced back and through.

3 SOLUTIONS

3.1 Data Accuracy and Consistency

Audit departments' data quality and integrity risks can

be greatly minimized by introducing an effective data

governance framework. The framework details how

data is captured, stored, processed, and shared and it

should be based on sound data management policies

and standards. In the audit procedure, a data

governance framework not only secures the

consistency and reliability of these systems'

performance but also prevents the occurrence of

wrong judgments that may arise from the data quality

issues. Besides, data stewards must frequently

perform data quality assessments and audits to verify

outdated, inaccurate, or incomplete record entries to

ensure data accuracy. The data governance machine

learning model has the potential to produce high-

quality input data due to the internal data cleansing

processes enforced by the audit departments, in turn

enabling highly reliable and accurate audit results.

Before the auditors use the machine learning

models, clean and prepare the data they receive using

advanced techniques. It can entail the use of

automating tools for detecting and correcting

inconsistency, errors, and problems with outliers. For

example, such algorithms can be applied during

financial transaction audits to discover potential

erroneous entries. Analyzing and processing lots of

data before exposing them to machine learning

models can eliminate influences from the data-related

biases and inaccuracies. This clearly demonstrates,

therefore, that the machine learning systems provide

with accurate and reliable insights and results during

audits, adding on the effectiveness of audits.

The range and detail of audits can be broadened

by gathering data from different sources and

integrating and verifying the collected information.

This includes incorporating data from various in-

house sectors such as finance, sales, and inventory

systems, as well as external data consisting of market

trends, economic indicators, and competitor

information. By including such extensive data into

the audit process, machine learning technology can

analyze the vast patterns and trends and hence

increase the precision of anomaly detection and risk

assessment. Moreover, auditors ought to perform

cross-validation on integrated data also because of its

verification and consistency, which is an amplifier of

the credibility of accounts.

3.2 Clarity and Explainability of Model

By integrating explainable AI like LIME or SHAP

into the pre-audit process, it becomes easier for the

auditors to scrutinize the developed system and obtain

consistent outputs. These technologies are useful not

only to explain how machine learning models reach

their decisions, being probably able to verify and

validate the correctness of outputs, and also to

auditors to check that everything stays conformant.

This clear connectedness is indispensable for the

auditors to check whether the models facilitate the

desired outcomes and subsequently precise their

results to stakeholders.

Machine Learning in Auditing: Problems, Solutions, Guidelines, Future Directions

499

Establish total model building documentation for

machine learning used in audit systems. This

document must contain the model's architecture,

assumptions, input data, output data, and procedures

on decision making. Furthermore, it is in the interest

of trust and reliability to record model inputs and

predictions in order to assist auditors in the

improvement of audit results.

Leo should create machine learning models that

are specifically designed for auditing, where

simplicity is an absolute necessity. This may involve

employing less complex approaches such as decision

trees or linear regression, for example, which are

straightforward as opposed to intricate deep networks.

While these models have slightly lower predictability,

they can still give you important explainable audits.

This can be related to understanding the decisions

made for compliance and accountability.

3.3 Model Overtraining and Auditors

Broadening of Model Use

Employ k-fold cross-validation or other cross-

validation methodologies to set the model’s

performance under different segments of the data to

test how well it performs on new data. This

methodology gives a better picture of underfitting by

analyzing consistency in validating models across

various data splits. The cross-validation method is, in

fact, a quite rigorous method of checking the

performance of the model, that is, its ability to

generalize to unseen data. Since it is very crucial for

the audit outcomes trustworthiness, it is widely used

in practice.

Integrate L1 (Lasso) and L2 (Ridge) regression

models for enforcing penalties on excessively

complex ones, which are also more likely to overfit.

Also, in tree-based classifiers, which are often

important activity in auditing, pruning is applied to

eradicate branches with unimportant impact or

unreasonable numbers based on noise in training data.

By means of these techniques, the model is simplified

and thus it is possible to generalize it for the new data

more easily.

All audit abusive scenarios should be considered

by making audit training data convenient and

representative of the different realities we see in our

work. This can involve increasing the training dataset

by creating synthetic data similar to that of infrequent

or unusual cases. The model has been trained over a

broad spectrum of examples, thereby it is more likely

to be generalizing instead of building a model for a

specific number of auditing contexts, with

considerable risk posed by overfitting.

3.4 Legal Compliance and Ethics

Regulations

Provide a range of audit-specific compliance

professionals, such as data privacy experts and ethics

officers, and ensure machines are appropriately

integrated into auditing processes. The desired

committees have the presence of lawyers, data

privacy professionals, auditors, and ethicists. Theirs

would be to develop policies and evaluate processes

ensuring that the use of machine learning models

abide by the relevant laws and regulations. Otherwise,

individuals who abuse data in processing and/or

usage without the right processes end up

compromising the audit process. Thereby, it is

unethical or not allowed according to the set

standards.

The demand for a transparent and auditable

system of machine learning in development and

deployment should be established. This may require

bringing the decision-making processes, data sources,

and data-utilization methods of these models into

clear light and publicly accessible. Transparency of

all algorithms and data processing routines

documentation not only develops confidence but also

creates an audit journal to investigate data privacy

and ethical problems.

The continuous assessment of ethical and

compliance of machine learning systems in auditing

ought to be done. These audits need to make sure that

the models are up to standard with the latest

regulatory requirements and ethical guidelines for

auditing. Areas of concern may include data use,

fairness of models, including “model bias” that make

some of these decisions subject to scrutiny, and

identification of possible threats. Regular audits

contribute to the immediate diagnosis of any

identified problems without fuelling severe

consequences. This, therefore, is used to confirm that

the use of machine learning in auditing is always

compliant and textually sound.

3.5 Knowledge and Experience of

Apprentices

Establish specific training programs on machine

learning, which are mainly oriented towards auditing.

This part should include basic machine learning

concepts, data analysis methods, as well as practical

functionalities such as anomaly detection, risks

assessment, and trend analysis in audit data. Real-life

scenarios may be carried through in case studies that

can exemplify how machine learning can be of

benefit in the auditor's day-be-day job. In this regard,

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence

500

the fact that they will have practical experience in the

field does not only help these auditors to grasp

machine learning concepts but also to be able to use

them in their works.

Generate and launch audit management tools and

systems that bring machine learning capabilities, thus

allowing the auditors to use these technologies. This

tool can reduce the burden on the auditors by

automating data processing, anomaly detection, and

risk identification tasks, thereby letting auditors to put

maximum emphasis on their judgment and decision

making. In this, the tools ought to have intuitive user

interfaces and offer detailed assistance and support to

help the auditors sooner adapt themselves to the

technologies and manage to acquire mastery of the

new techniques.

Get a machine learning support team in-house

dedicated to auditors or utilize the external machine

learning consultants to provide assistance with

complicated technical problems. This team can act as

a helpdesk to help auditors with technical problems

regarding the application of machine learning or with

complex data analysis tasks. Consequently, auditors

will be able to internalize and apply the machine

learning techniques while assuring the correctness

and quality of the audit procedures.

4 BEST PRACTICES

4.1 Continuous Learning and

Development

To give auditors a chance to get acquainted with the

possibilities of machine learning tools, companies

should regularly hold training events to introduce the

fundamentals of ML, recent advances, and their use

in auditing. These training courses can solve practical

problems. For example, auditors can receive practical

case study examples and will be able to apply what

was learned in the training room to the real world that

awaits this concept in its preparation. This practical

method provides that auditors comprehend the

concepts of machine learning and getting skills to use

them by auditors accordingly.

4.2 Form Cross-functional Teams

It is better to involve scientists, audit experts, and IT

professional within interdisciplinary teams to audit

those machine learning models. These groups may

work together to obtain and fine-tune models that are

scientific yet still meet the requirements of auditing.

While working in a teamwork manner, the audit

professionals tend to get a better insight into the audit

specifications and hence design appropriate technical

solutions. Such an inclusive approach always ensures

that the developed models are not only technically

good but are also practically important for auditing.

4.3 Develop Agile Development

Methodologies for Businesses

Adopting Agile development techniques is the best

way to apply machine learning in auditing. This

approach lets teams build, analyse, and retest their

models quickly, uniquely modifying them based on

the feedback they get. Agile methodology supports

the flexibility and correctness of those models and

aids quick detection of possible disruptions that could

arise and cause serious losses. Up to date, by using

iterative development, organization can gradually

improve design of their models according to the

changes in audits processes.

4.4 Improve Data Management and

Governance

The quality of data and its management bore keen

importance for the use of machine learning is a

successful audit. We must come up with a data

management and governance model that addresses

these critical dimensions to enhance the data

reliability, validity, and completeness. The entire

process of data generation, storing, running, and

sharing should obey strict standards, protocols, and

guidelines to help the machine learning applications

along with the efficient and reliable ones. As a

process, data governance not only improves the audit

quality but also allows for compliance with rules

governing data protection and privacy.

4.5 Model Validation and Continual

Improvement

Application of the regular model validation and

evaluation methods is a primal stage keeping the

quality in check of the machine learning models as

well as the accuracy of the audit results. Through

ongoing observation and feedback loops,

organizations become capable of quickly spotting

aberrations in the models, which inevitably leads to

their consistent and optimal performance. Regular

validation keeps the models in line with auditing

standards and practices upgrade, making them the

tool which remain effective in long term.

Machine Learning in Auditing: Problems, Solutions, Guidelines, Future Directions

501

5 FUTURE DIRECTIONS

5.1 Broadening the Field of

Interpretability and Transparency

Interpretability as well as transparency are both

critical in machine learning of audits as the auditors

need to comprehend and explain the reasoning

process of machine learning models to ensure their

fair and accurate results. Although complex models,

such as deep learning, achieve the aforementioned

results through their black-box nature, full knowledge

of model processes is complex and difficult for

auditors. The next step of building interpretable ML

methods should become the key goal of modern

research, both in academic and industrial

environments. In addition to the current models like

LIME (Local Interpretable Model-agnostic

Explanations) and SHAP (SHapley Additive

exPlanations), the other important aspect is to see

how we can leverage the neural networks and deep

learning models and make them more transparent.

There're different ways to achieve that, these include

putting forward user-friendly model architectures or

introducing interpretability as an objective function

during model development. Implementing such

techniques would serve not only in the proper

understanding of the output of the models but also the

increasing of the credibility of theses so that the audit

findings can be relied on and accepted.

5.2 Integration with Multimodal Data

Auditing focus relies too much on financial

dimensions in reality and does not harness the

opportunity of unstructured data. Next-generation

audit technologies must be multifunctional, hence

should be able to handle and combine multiple data

types, such as text, images, and audio. This is made

possible by multimodal machine learning models,

which allow the auditors to observe and learn from

studying data from various scope. One option could

be to use NLP techniques to perform an analysis of

the internal communication records (such as email or

chat logs) that are stored within a company to help out

in the identification of fraud or any other compliance

violations. On the contrary, computer vision

techniques can be used to reveal discrepancies and

anomalies that might be present in scanned contracts

and documents. It will vastly improve the uniformity

and correctness of the audit process to consolidate

into one the complementary types of data that

insightfully goes together in audit results.

5.3 Live Auditing and Continuous

Monitoring

With the operations environment of businesses

becoming much more sophisticated and evolving

rapidly, instant auditing with constant surveillance

seems to be the trend. Consequently, such a viewpoint

requires machine learning models to ingest and

comprehend data streams from a multitude of sources,

with the goal of immediate response to anomalies and

risks, which may appear at the same time. The

resultant effect is that such real-time data stream

processing algorithms and systems architectures that

are equipped with the ability to continuously monitor

business processes and detect anomalies based on an

established criterion or model will be developed.

Taking note of the potential for financial crimes at

that moment, real-time tracking of financial

transactions and account activity could provide notice

about the existence of any possible fraud or financial

crime and prompt the undertaking of corrective

actions. For such an accomplishment, studies would

need to be extended to not only the development of

ML models, but also resource management that

involves computing, communication, and system

integration.

5.4 Firming up Ethics and Regulatory

Compliance

Machine learning use in auditing undoubtedly brings

the challenges in data privacy and ethics with it.

Future research should concentrate on how to make

the integration of data utilization and analysis not

only effectively but also maintaining the data privacy.

The algorithms of privacy maintenance, for instance,

differential privacy, can lessen the chance of

disclosure of sensitive information about some

personal parts during the processing of the data.

Additionally, the development of data protection

regulations such as GDPR and CCPA pushes

organizations into designing compliance algorithms

to make sure that models’ operations do adhere with

laws of the land. Another central issue is algorithmic

bias; machine learning models may unconsciously

capture the biases that were present in the training

data. Thus, a future research focus should be on

algorithms which are developed to detect and

minimize bias, thus ensuring models make impartial

and fair decisions. The features will not only increase

the acceptability of models in the legal realm and

social but also help in enhancing their applicability in

the auditing field.

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence

502

6 CONCLUSIONS

In a business environment that is so varied and

competitive, auditing is still valuable in terms of

book-keeping, adherence to laws, and financial

system stability. The inclusion of machine learning

(ML) has brought with it a slew of benefits in the form

of improved accuracy and processing times in audit

routines, such as fraud detection, risk profiling, and

automated auditing reports, to name a few. However,

despite the convenience, the application of ML in the

area of auditing still has some issues that need to be

addressed. These problems are features of data

quality, model transparency, overfitting, and

regulatory compliance. Also, ethical concerns in

combination such as data ownership and fairness of

algorithms have to be dealt with prudence. Instead of

looking for the benefits of implementing ML in

auditing, data governance frameworks should be

reinforced, model interpretability improved, and

training provided for auditors. Tackling these

problems will enable the fair usage of ML in auditing,

which is bettering audit quality in addition to keeping

the right ethical and legality standards.

REFERENCES

Adler, P., et al. (2018). Auditing black-box models for

indirect influence. Knowledge and Information

Systems, 54, 95-122.

Cao, L. (2017). Data science: a comprehensive overview.

ACM Computing Surveys (CSUR), 50(3), 1-42.

Carvalho, D. V., et al. (2019). Machine learning

interpretability: A survey on methods and metrics.

Electronics, 8(8), 832.

Criado, N., et al. (2021). Attesting Digital Discrimination

Using Norms.

Fasterling, B. (2012). Development of norms through

compliance disclosure. Journal of Business Ethics, 106,

73-87.

Hassan, G. M. (2023). Influence Of Corporate Governance

Practises on Service Delivery Among County

Governments In Kenya. KCA University,

Ilori, O., et al. (2024). Advanced data analytics in internal

audits: A conceptual framework for comprehensive risk

assessment and fraud detection. Finance & Accounting

Research Journal, 6(6), 931-952.

Karpoff, J. M., et al. (2012). A critical analysis of databases

used in financial misconduct research. Mays Business

School Research Paper, 73(2012), 2012-2011.

Knechel, W. R. (2013). Do auditing standards matter?

Current Issues in Auditing, 7(2), A1-A16.

Odeyemi, O., et al. (2024). The role of AI in transforming

auditing practices: A global perspective review. World

Journal of Advanced Research and Reviews, 21(2),

359-370.

Stefanov, E., et al. (2012) “Iris.” Proceedings of the 28th

Annual Computer Security Applications Conference on

- ACSAC ’12.

Song, Q. C., et al. (2021). Making sense of model

generalizability: A tutorial on cross-validation in R and

Shiny. Advances in Methods and Practices in

Psychological Science, 4(1), 2515245920947067.

Machine Learning in Auditing: Problems, Solutions, Guidelines, Future Directions

503