Enhancing Aviation Safety Analysis in MROs:

A Complex Emergent Model with a Predictive Approach

Victoria Grech

and Joseph Paul Zammit

Department of Industrial & Manufacturing Engineering, University of Malta, Msida, Malta

Keywords: Complex, Data Mining, Emergent, Machine Learning, MRO, Non-Linear, Occurrence Reporting, Predictive,

Safety Causation Modelling, Safety Management Systems.

Abstract: This study explores the aviation industry's shift from reactive and proactive safety strategies towards

predictive safety management, focusing on Maintenance, Repair and Overhaul (MRO) operations. It

introduces a novel complex emergent safety model designed to integrate predictive analytics into existing

Safety Management Systems (SM) via occurrence reporting data. Moving beyond traditional linear causation

models, the proposed framework leverages machine learning and data mining techniques to identify hazards

and assess risks, thereby reducing the frequency and severity of incidents and minimising maintenance

disruptions. Using the DMADOV methodology, the study aims to extract actionable insights from unexploited

safety data, despite challenges such as data quality variations and the stochastic nature of safety. Ultimately,

this research advocates for a unified, AI-driven approach to enhance safety capabilities across the aviation

industry.

1 INTRODUCTION

Highly technological and risk systems in high

reliability industries such as aviation are becoming

increasingly complex, raising the potential for

catastrophic consequences when failures occur.

(Qureshi, 2008) Accidents in aviation, as per ICAO

Doc 9156 (ICAO, 1987), are defined as events leading

to serious injuries, significant aircraft damage, or the

aircraft being missing or inaccessible, during

passenger embarkment and disembarkment. Incidents,

on the other hand, are occurrences that could impact

flight safety, ranging from aircraft operations,

technical issues, to interactions with air navigation

services and environmental factors. The distinction

between accidents and incidents primarily lies in their

severity and impact. (ICAO, 1987), (European

Parliament and Council, 2015) Understanding both

accidents and incidents is crucial, leading to the

development of various safety causation models,

approaches and methodologies. This evolving safety

thinking underlines the industry’s commitment

towards safety reassurance, especially when

https://orcid.org/0009-0008-8515-3829

https://orcid.org/0000-0002-9271-9682

considering how relatively young the aviation industry

is. (J. J. a Stoop & Kahan, 2005).

Safety causation models are theoretical

conceptual frameworks that generate a reasoning of

occurrences. They try to explain how and why

accidents occur. Accident modelling can be traced to

as early as the 1920s. Upon such safety causation

models, accident investigations are used to describe

and explain the occurrences. (HaSPA (Health and

Safety Professionals Alliance), 2012; Reason, 1990)

This reactive method did not assist in identifying the

problem and subsequently could not prevent a similar

incident. In the 1950s, aviation safety investigations

shifted from identifying technical factors, to human

factors and then moved towards organisational factors

to try to understand and solve accidents. Initially, as

the aircraft was considered as a complex technological

marvel, the main factor for failure was equipment.

Then, as technology became more reliable, the focus

shifted towards human factors. This Era brought about

the concept of Crew Resource Management and was

solely focused on the individual. Towards the 1990s,

this progressed into considering the operational

context of a complex environment within an

340

Grech, V. and Zammit, J. P.

Enhancing Aviation Safety Analysis in MROs: A Complex Emergent Model with a Predictive Approach.

DOI: 10.5220/0013667400004000

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2025) - Volume 2: KEOD and KMIS, pages

340-348

ISBN: 978-989-758-769-6; ISSN: 2184-3228

organisation. This systemic perspective led into

providing the basis for the development of a Safety

Management System, SMS. (Reason, 1990)

SMS emerged as a central pillar of modern

aviation safety strategy, placing greater emphasis on

regulations, organisations and real-time decision-

making for safety improvements. Defined by ICAO,

SMS encompasses managing organisational

structures, responsibilities, and procedures to enhance

safety through occurrence reporting. (Gerede, 2015;

Yeun et al., 2014) Moving beyond traditional

prescriptive approaches, SMS adopts a realistic view

of the world, encouraging a paradigm shift towards

proactive and predictive safety measures. (Gerede,

2015) EU Regulation 376/2014 reflects this evolution

by advocating for the integration of proactive methods

with reactive systems for more effective safety

improvements. (European Commission, 2014) SMS

fosters a safety culture by merging reactive, proactive,

and predictive strategies across organisational levels,

showing early benefits and driving continuous safety

advancements. Reactive methods focus on immediate

incident response and mitigation, while proactive and

predictive strategies aim to prevent future incidents by

identifying and anticipating risks, creating a

comprehensive approach to organisational safety

management. These are illustrated as per Figure 1.

(Safety Management System | Federal Aviation

Administration, n.d.)

Figure 1: Reactive-Proactive-Predictive methodologies.

Regulations and traditional safety management

practices aim to incorporate corrective, preventive,

and predictive methodologies into SMS, with a focus

on corrective measures and an encouragement of

preventive actions. Following EU Regulation

376/2014 (European Commission, 2014), one of the

key tools for SMS is occurrence reporting or safety

investigations (Elkhweldi & Elmabrouk, 2017), which

involve fault analysis for corrective actions and cause

analysis for preventive measures. Stoop and Dekker

(J. Stoop & Dekker, 2012) question the proactiveness

of safety investigations, highlighting the importance

of feedback from real-world data for insights into

complex systems. This knowledge is crucial for future

designs and strategies, making safety investigations a

proactive element that complements other safety

improvement strategies. However, even though the

predictive phase of SMS is recognised it is yet to be

defined, clarified, encouraged and enforced by

regulations and authorities to organisations.

The predictability of incidents, applied in a

maintenance, repair and overhaul (MRO), which

makes up just one aspect of the aviation industry,

presents a complex and critical challenge. This is due

to the multifaceted nature of aviation systems, the

variability in operational environments, and the

stringent safety standards required. The primary

challenge lies in the development and limitations of

current safety causation models that are reductionist,

linear and resultant. The illusion of containment or

preventing ‘losses’ gives the impression that incidents

and accidents alike can be controlled. Many models

have a Newtonian Cartesian ideology, that the incident

or accident can always be broken down. This leads to

a hunt for a broken component. Currently, with no

universally accepted model (Grant et al., 2018), the

pessimistic conclusion would be that the models are

not scientific enough, practical enough, not specific

enough nor holistic enough to fully understand how

incidents occur. (Hovden et al., 2010) In light of the

(r)evolution of many safety causation models,(HaSPA

(Health and Safety Professionals Alliance), 2012)

unfortunately, their outlook in the current predictive

approaches remains limited. The continuous growth

and advancements in all aspects of the aviation

industry necessitate further developments, which

include the way accidents are viewed and their

methodologies applied (Amankwah-amoah, 2021).

A complex emergent model that integrates all

three rationales – reactive, proactive and predictive is

required. This research aims to lay the foundation for

complex, non-linear safety thinking in both incident

and accident investigations. By integrating a

predictive-probabilistic analysis approach into the

existing SMS this approach will enhance the capacity

to foresee and mitigate safety risks. It aims to

statistically reduce the frequency and severity, while

minimising disruptions to maintenance operations.

(Bartulović & Steiner, 2023).

2 LITERATURE REVIEW

The evolution of safety causation models, from

Heinrich's 1931 Domino theory to Hollnagel’s 2012

Functional Resonance Analysis Method (FRAM),

reflects a shift through three generations of human

Enhancing Aviation Safety Analysis in MROs: A Complex Emergent Model with a Predictive Approach

341

error modelling (Katsakiori et al., 2009),

incorporating human factors and systemic approaches

like Reason’s Swiss cheese model. Despite challenges

in application and interpretation, these models have

progressively addressed the complexity of safety

management, culminating in the adoption of complex

non-linear models in the early 2000s, such as

Leveson’s STAMP and FRAM, to tackle the dynamic

aspects of safety. (Hovden et al., 2010) This literature

review critically assesses these models for their

principles, strengths, and weaknesses, aiming to

promote a shared understanding of accidents and

support the development of preventative strategies.

(Hovden et al., 2010) By analysing the interaction

among various factors within these models and

highlighting their unique features, the review

advocates for a comprehensive model that addresses

the complexity and emergence of incidents,

streamlining the evolution and critical examination of

safety causation models and their application in

enhancing system resilience and safety management.

The Swiss Cheese model was introduced by James

Reason and conceptualizes the idea of multiple layers

of defence against accidents in complex systems. Each

layer of defence has potential flaws, represented as

holes in slices of Swiss cheese. The alignment of these

holes can lead to a trajectory of accident opportunity,

allowing hazards to materialize into losses. The

model's strength lies in its visual simplicity and its

emphasis on systemic flaws rather than individual

error. However, its limitation is the linear and static

representation of accident causation, overlooking the

dynamic interactions within systems and the non-

linear nature of complex failures.(Reason, 1990)

While some authors (Dekker, 2002; Maurino, 2001;

Shappell & Wiegmann, 2000), considered the model

to be too generic and underspecified pinning the

model as a representation which lacks the tools to

implement the metaphor of cheese’s slices and holes.

This leaves practitioners making their own

interpretation and adaptation. While Luxhøj and

Kauffeld (Kauffeld, 2003), think that this is a risk

which makes the model impractical, this interpretive

flexibility suits particularly well the SCM. (Larouzee

& Le Coze, 2020) In fact, in 2000, a simplified version

of the SCM was published in the British Medical

Journal (BMJ)(Reason, 2000) making an impact in

another high-risk industry.

Charles Perrow's Normal Accident Theory (NAT)

suggests that accidents are a natural outcome in

complex, tightly coupled systems due to the

unpredictable and unmanageable nature of their

interactions. It highlights the intrinsic risks within

high-tech environments, suggesting that the

complexity of these systems renders accidents

inevitable, hence ‘normal’. As this theory recognizes

accidents as such, it risks diminishing the emphasis on

proactive risk management (Charles, 1999).

FRAM,

developed by Erik Hollnagel, focuses on how

variability in normal system performance can lead to

accidents through unexpected interactions. Unlike

linear models, FRAM addresses the complexity and

non-linear interactions within systems, offering a

more dynamic approach to understanding accident

causation. This is possible because through the

concept of resonance, a small change in one part could

amplify and cascade into larger ones. Later, it was also

mentioned and explained as the butterfly effect by

Dekker. (Dekker, 2017) Moreover, the FRAM’s

strength lies in its ability to model complex processes

and their variabilities. It provides an analysis method

by defining a system in terms of functions, which

represent activities that people perform, whereas each

function can be defined by six aspects; the time

allocated, input items that are processed,

preconditions that trigger the task, the way the

function is controlled, the resources that are consumed

to process items and the output as a result of the

function. However, its application can be challenging

due to the need for in-depth understanding of system

variabilities and interactions. It requires many

resources which would require time to retrieve. In

addition, the complex connectivity of web could lead

to an infinite number of possibilities making it

difficult to predict how variabilities across different

functions may couple and resonate. (Erik, 2012)

The Systems-Theoretic Accident Model and

Processes (STAMP), Proposed by Nancy Leveson,

shifts from event-based to constraint-based

approaches in accident analysis. It views accidents as

a result of inadequate control or enforcement of safety

constraints within a socio-technical system. STAMP's

strength lies in its comprehensive approach,

incorporating technical, human, and organizational

factors. However, its broad scope can make it complex

to implement and require significant effort to identify

and model relevant constraints.(Leveson, 2004) This

model faces challenges that limit its widespread

adoption when compared to other models because it

requires a deep understanding of system theory which

is not synonymous with systems engineering and

hence, requires specific education or training. It is also

challenging to apply real world application of STAMP

to complex systems, hence, make it challenging for

practitioners to apply it to their specific contexts. Its

emphasis on enforcing constraints on system

behaviour offers a novel approach to safety, which,

with increased training and awareness, could see

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

342

broader adoption across various industries. (Leveson,

2004, 2012; Underwood & Waterson, 2012)

However, Roelen, Lin and hale (Roelen et al., 2011)

suggest that this model is neither embraced by the

safety community nor broadly recognised as a

significant influencing accident model in the overall

field of safety management because Leveson’s model

does not integrate well with the dominating methods

of collecting and analysing safety data. Hence, making

event chain models such as the SCM more favourable.

Dekker's Drift into Failure theory proposes that

accidents result from systemic drifts into failure,

where everyday decisions and actions, although

seemingly rational at the time, cumulatively lead to a

system's degradation and eventual failure. This model

emphasizes the complexity of socio-technical systems

and the non-linear, emergent properties of system

failures. Its strength lies in its focus on the systemic

and emergent nature of failures, challenging the blame

culture by highlighting the role of system design and

decision-making processes. (Dekker, 2017) While

Dekker provides a theory and tries to progress towards

a complex and emergent kind of rationale, this theory

does not provide a safety causation model that fully

explains how an incident, or an accident occurs.

Therefore, a limitation of this theory could be in the

identification and measurement of drift, which is

subtle and evolves over time.

The reviewed models contribute valuable insights

into accident causation, highlighting the roles of

system complexity, human factors, and organizational

processes. However, there exists a gap in integrating

these perspectives into a unified model that can

accommodate the emergent, unpredictable nature of

complex system failures. Current models either focus

on linear causation, system components in isolation,

or fail to fully address the dynamic interactions and

adaptability of complex socio-technical systems.

There is a need for a model that not only incorporates

the strengths of existing models, such as the graphical

visualisation, systemic nature and flexible

interpretation of the SCM, the dynamic understanding

of FRAM, the normalisation of unstable systems of

NAT, and the drift concept with an understanding of

non-linear, complex emergent properties from

Dekker; but also offers practical tools for identifying

and mitigating emergent risks in real-time to keep up

with the ever growing technologies.

For the enhancement of safety investigation

analysis, this research aims at developing an

integrated safety causation model (Grant et al., 2018)

that embraces complexity, emergence, and the non-

linear dynamics of socio-technical systems. Such a

model would provide a more holistic and adaptable

framework for understanding and preventing

accidents in an increasingly complex and

interconnected world. Alongside this integrated

model, an analysis will also be developed. This would

reflect the novel’s notion of non-linearity, complexity

and emergence in contrast to previously mentioned

models with their own analysis methods or rationale.

3 METHODOLOGY

The DMADOV (Define, Measure, Analyse, Design,

Optimise, Verify) methodology (Pyzdek, 2017) will

be applied with the goal of designing and developing

a predictive-probabilistic, integrated data-based

analysis approach. This approach will complement the

non-linear, complex, and emergent safety causation

model, serving as a practical tool. Specifically, it aims

to identify potential areas of emergent risks in

maintenance environments by conducting a thorough

analysis of already existing unexploited safety data.

In the define phase, the MRO organisation’s

current SMS will be examined, focusing on the safety

investigation process and occurrence report inputting.

This involves a thorough review of the existing

process for investigating safety incidents, identifying

its strengths and limitations. Written occurrence

reports, generated from incident investigations, will be

looked into to grasp their significance into aviation

safety. This will support in formulating problem

statements considering the inputting fields and data

available. By defining these reports’ qualitative and

quantitative nature of data, a deeper understanding of

the involved processes and the significance of report

fields will be gained. This understanding would

potentially indicate the specific relationships that need

to be measured and subsequently analysed.

Once the investigative process as well as the

occurrence reports’ fields and criteria would be

understood, the data available will be explored into

setting attributes from the defined problem statement.

This would be the measure phase. Identifying

relationships from the investigative process and

occurrence reports’ criteria and fields will give an

indication of specific attributes and categories to

extract. Following the identification of such attributes

and categories, which must also be inline with the

defined problem, the type of data must be studied

further. This will lead into extracting the data,

cleaning it and refining its quality. The selection of

attributes and categories of data, together with its

filtration and refinement process are of crucial

importance to provide a clear as-is assessment of the

reporting data. (Huan & Hiroshi, 2007)

Enhancing Aviation Safety Analysis in MROs: A Complex Emergent Model with a Predictive Approach

343

In the analysis phase, techniques and tools for data

extraction are chosen based on predefined criteria,

often requiring data transformation to meet analysis

needs. Artificial Intelligence (AI), Data Mining (DM),

and Machine Learning (ML) are key areas in this

process. (Huan & Hiroshi, 2007) AI is about creating

systems that perform tasks requiring human

intelligence, like speech recognition and decision-

making. ML, a subset of AI, focuses on developing

algorithms that enable computers to learn from data

and improve over time without being explicitly

programmed. This is crucial for AI systems to adapt

and perform complex tasks accurately. DM discovers

patterns in large data sets using ML, statistics, and

database techniques, converting data into a usable

structure. (Han et al., 2011) While DM aims at

exploratory analysis to uncover new insights, ML

focuses on using these insights for predictions or

classifications. Current studies employ DM to uncover

patterns and knowledge in data, and ML to learn and

make predictions, illustrating ML's versatility across

various industries such as healthcare and technology.

(Revolutionizing Healthcare Industry with Machine

Learning., n.d.) DM supports ML by providing

algorithms for systematic data analysis, crucial for

understanding customer behaviour, optimising

efficiency, and identifying risks or opportunities (A.

B.Arockia & S. Appavu, 2013). Together, ML and

DM enable organisations to automate data analysis,

enhance prediction accuracy, and make informed

decisions, essential in data-intensive fields like

aviation where safety and data volume are critical.

Data mining plays a pivotal role in predictive analysis

by employing a range of techniques and tools to

extract actionable insights from datasets. Key tools

such as WEKA, RapidMiner, KNIME, and Python

Libraries support various DM techniques including

classification, clustering, regression, association rules,

and anomaly detection. Prominent classification

methods like decision trees (ID3, C4.5, CART),

support vector machines (SVM), naive Bayes (NB),

and K-nearest neighbours (KNN) have been

extensively utilized. For instance, decision tree

induction is celebrated for its efficacy in pattern

classification by Han and Kamber (Han et al., 2011),

while NB classifiers and SVM have been applied for

their probabilistic and discriminative capabilities in

aviation safety analysis respectively. (Narasimha &

Devi, 2011), (Han et al., 2011) KNN, known for its

simplicity and effectiveness, along with decision trees,

has contributed to understanding complex data

structures in aviation incidents and forecasting models

(A. B.Arockia & S. Appavu, 2013), (Gürbüz et al.,

2009), (Bineid & Fielding, 2003).

In this analysis phase, the key tools and DM

classification techniques will be selected based on the

previous phases, highlighting the significance of

choosing the right combination of techniques and

tools based on specific analysis objectives, essential

for deriving meaningful insights and enhancing

aviation safety.

In the following phase, machine learning methods

and algorithms will be applied through data mining

tools and techniques to study relationships and

recognise existing trends. This will lead towards

evoking strategies to compare criteria, fields and

records in the design phase. Utilising a machine

learning platform together with the deployment of

functions, algorithms and selected methods, a design

procedure will be established. The selection of

different functions to assess performance levels and

their impacts will necessitate redesigns. These

redesigns will make part of the optimize phase.

Through the iteration of the design and optimise

phases, the previously defined data and relationships

will undergo verification for reliability purposes. This

methodology will enable the validation of the

predictive-probabilistic approach, in the final verify

phase.

After the literature review of the safety causation

models aimed at paving the way for the development

of a complex emergent model, the DMADOV

methodology and its phases were discussed for the

development of a predictive-probabilistic analysis

approach, which is defined further in Figure 2.

Ultimately, through this analysis approach, applied on

specified criteria and attributes of safety occurrence

reporting data, it would also support the complex

emergent safety causation model in understanding the

occurrence of incidents.

Figure 2: DMADOV phases.

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

344

4 POTENTIAL IMPACT

The study embarks on an innovative journey to

develop a complex emergent safety causation model,

leveraging the power of machine learning algorithms

and methods to foster a predictive-probabilistic

approach. Utilising data from MRO operations, this

research underlines the importance of advancing

safety thinking in high-risk industries, notably

aviation. This progressive exploration aims to unravel

and navigate through the intricacies and emergent

properties of safety data, promoting a more profound

understanding of underlying risks.

Based upon the results of Mathur et al (Mathur et

al., 2017), it was analysed that maintenance of the

aircraft would significantly reduce the risk of

accidents. Therefore, by integrating innovative data

mining applications and employing pragmatic

predictive-probabilistic strategies through advanced

tools, techniques, and methodologies, the study seeks

to indicate potential risk incidents before they

manifest in an MRO environment. This proactive

detection, rooted in the meticulous analysis of safety

occurrence reporting data, would substantiate positive

impacts. Beyond the mere avoidance of adverse

outcomes, this approach signifies a leap towards

ensuring safer working environments and enhancing

operational efficiency. The anticipated reduction in

downtime and interruptions from incidents paves the

way for heightened productivity, as resources

dedicated to managing or responding to incidents are

optimised. Furthermore, this efficiency translates into

cost savings, encompassing not only operational costs

but also expenses related to repairs, legal fees, and

more. (Jardine & Tsang, 2013) The ripple effects of

diminished incidents extend to encouraging trust

among employees and customers, fortifying a

reputation for reliability and safety. Such a reputation

could serve as a formidable competitive edge, drawing

more business and elevating customer satisfaction.

(Doorley & Garcia, 2007) The insights derived from

this study are expected to shed light on risk areas and

vulnerabilities, guiding strategic decisions and

preventive actions. This knowledge empowers

organisations to learn from incidents that occur,

nurturing a culture of learning and adaptation. It also

encourages the adoption of best practices, steering

continuous improvement across the board.

5 LIMITATIONS

In developing a predictive, machine learning-driven

models, this study faces numerous challenges

inherent to the domain of safety performance. The

stochastic nature of safety, characterized by complex

interactions and unpredictable events, complicates

precise quantification and modelling. (J. Stoop &

Dekker, 2012) Transitioning from reactive to

proactive safety strategies in MRO operations further

adds to the complexity, highlighting the importance

of training, awareness, and fostering a just culture.

(Gerede, 2015), (Phimster et al., 2004) Notably, on

the latter, Gerede (Gerede, 2015), continues by

mentioning the challenging barriers of 'just culture'

that pose on Safety Management Systems, impeding

effective reporting, learning, and predictive tool

enhancement. This study employs data mining tools

and techniques to model safety occurrences’ defined

queries facing several data limitations within MRO

operations. The inherent complexity and abstract

nature of safety models necessitate significant

simplification for practical application, challenging

due to data quality issues like incompleteness and

inconsistency. Moreover, the substantial resources

required for processing large datasets introduce

technological and computational constraints.

(Arockia Christopher & Appavu Alias Balamurugan,

2014) The probabilistic nature of safety incidents

adds another layer of uncertainty, affecting model

reliability and necessitating extensive validation on

extensive high-quality data. (Gao & Mavris, 2022)

Integration and standardisation of historical data

across reporting systems are hindered by technical

and regulatory barriers. Additionally, achieving a

balance in predictive models to minimise false

positives requires sophisticated algorithms capable of

adapting to new aviation technologies and failure

patterns. The application of Heinrich's pyramid

highlights the significance of unreported incidents in

the accuracy of safety models, underscoring the

challenge of capturing comprehensive data. (Nazeri

et al., 2008), (IATA - IATA Releases 2022 Airline

Safety Performance, n.d.) Finally, incorporating

human factors, including variability in maintenance

practices and the potential for human error,

introduces further complexity, emphasizing the need

for a nuanced approach to modelling safety in

aviation maintenance and operations.

To conclude, while this research pioneers a

predictive rationale for complex emergent safety

causation modelling and strives to apply data mining

tools and techniques, it faces various limitations

ranging from data quality, availability and

technological constraints to the unpredictable human

element. Addressing these challenges requires

extensive efforts to refine methodologies, enhance

Enhancing Aviation Safety Analysis in MROs: A Complex Emergent Model with a Predictive Approach

345

data management practices, and foster a culture that

supports proactive safety management.

6 FUTURE DIRECTION

Given the insights from Nazeri et al. (Nazeri et al.,

2008) regarding accident factors and considering the

projected growth in air transportation, we stand at a

critical juncture. The increasing complexity of air

traffic and fleet sizes, alongside the inherent

aggravation of conditions affecting accident factors,

suggests that accident rates could escalate if proactive

measures aren't taken. Current safety investigation

methodologies, despite their historical efficacy, face

criticism for obsolescence. This highlights the urgent

need for a paradigm shift in safety investigations,

emphasising data quality, proactive outcomes, and a

future-oriented mindset when conducting an

investigation (J. Stoop & Dekker, 2012), (A.

B.Arockia & S. Appavu, 2013).

A complex emergent safety causation model

incorporating effectively all three rationales will be

developed and formulated based upon key features

and characteristics identified from already existing

models and theories. These would include relatability

and flexibility while also catering for non-linearity,

recognise complex systems and consider emergent

properties. Therefore, path the way for complex non-

linear emergent safety thinking in incident and

accident investigations. An analytical predictive

approach will be formulated through the DMADOV

methodology whereas the qualitative and quantitative

data will be processed through the methodology’s

phases. Based on previously mentioned case studies,

data mining techniques and tools will be applied to

study relevant relationships categories and fields.

Following through with the analysis, it will continue

to fulfil the three rationales by integrating the

predictive analysis approach into the existing SMS.

The transition towards predictive risk management

denotes a promising direction in enhancing aviation

safety. Further potential impacts and limitations were

then discussed.

In conclusion, the path forward demands a

collaborative effort to embrace predictive models of

safety management. This paper represents a

conceptual first step, introducing a complex-

emergent safety causation framework for MROs.

While the model is currently theoretical, future work

will focus on its empirical validation using real-world

occurrence reporting data. This includes testing and

refining the framework using machine learning

methods to evaluate its practical impact on predicting

and mitigating safety risks. By harnessing data and

predictive analytics, risks could be anticipated to

foster a safer future for aviation. Potentially spreading

into other high-risk industries like nuclear power and

healthcare. (Nazeri et al., 2008) This transition, while

challenging, is essential for advancing safety culture,

offering a proactive approach for managing the

complexities of modern aviation and beyond.

REFERENCES

A. B.Arockia, C., & S. Appavu, A. B. (2013). Data mining

approaches for aircraft accidents prediction: An

empirical study on Turkey airline. 2013 IEEE

International Conference on Emerging Trends in

Computing, Communication and Nanotechnology,

ICE-CCN 2013, Iceccn, 739–745. https://doi.org/10.

1109/ICE-CCN.2013.6528602

Amankwah-amoah, J. (2021). COVID-19 pandemic and

innovation activities in the global airline industry: A

review. Environment International, January.

Arockia Christopher, A. B., & Appavu Alias Balamurugan,

S. (2014). Prediction of warning level in aircraft

accidents using classification techniques: An empirical

study. The Aeronautical Journal, 118(1206), 1217–

1223. https://doi.org/10.1007/978-81-322-1665-0_126

Bartulović, D., & Steiner, S. (2023). Conceptual Model of

Predictive Safety Management Methodology in

Aviation. Aerospace, 10(3). https://doi.org/10.3390/

aerospace10030268

Bineid, M., & Fielding, J. P. (2003). Development of a civil

aircraft dispatch reliability prediction methodology.

Aircraft Engineering and Aerospace Technology, 75(6),

588–594. https://doi.org/10.1108/00022660310503066

Charles, P. (1999). Normal Accidents: Living with High-risk

Technologies. Princeton University Press.

Dekker, S. W. A. (2002). The Field Guide to Human Error

Investigations. Ashgate. https://doi.org/10.4324/

9781315202778

Dekker, S. W. A. (2017). Drift into Failure: From Hunting

Broken Components to understanding complex systems.

In BMC Public Health (Vol. 5, Issue 1). CRC Press

Taylor and Francis Group.

Doorley, J., & Garcia, H. F. (2007). Reputation Management.

The Key to Successful Public Relations and Corporate

Communication. Routledge Taylor & Francis Group.

Elkhweldi, M. A., & Elmabrouk, S. K. (2017). Aviation Risk

Management Strategies. Proceedings of the 2015

International Conference on Industrial Engineering and

Operations Management Dubai, United Arab Emirates

(UAE), March 3 – 5, 2015, March 2014, 1–7.

Erik, H. (2012). FRAM: The Functional Resonance Analysis

Method: Modelling Complex Socio-Technical Systems.

https://doi.org/10.1201/9781315255071

European Commission. (2014). Regulation (EU) No

376/2014 of the European Parliament and of the Council

on the reporting, analysis and follow-up of occurrences

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

346

in civil aviation, amending Regulation (EU) No

996/2010 of the European Parliament and of the council.

Official Journal of the European Union, 18.

European Parliament and Council. (2015). Commission

Implementing Regulation (EU) No 2015/1018 of 29 June

2015 laying down a list classifying occurrences in civil

aviation to be mandatorily reported according to

Regulation (EU) No 376/2014 of the European

Parliament and of the Council. Official Journal of the

European Union, L163, 1–17.

Gao, Z., & Mavris, D. N. (2022). Statistics and Machine

Learning in Aviation Environmental Impact Analysis: A

Survey of Recent Progress. Aerospace, 9(12).

https://doi.org/10.3390/aerospace9120750

Gerede, E. (2015). A study of challenges to the success of the

safety management system in aircraft maintenance

organizations in Turkey. Safety Science, 73, 106–116.

https://doi.org/10.1016/j.ssci.2014.11.013

Grant, E., Salmon, P. M., Stevens, N. J., Goode, N., & Read,

G. J. (2018). Back to the future: What do accident

causation models tell us about accident prediction? Safety

Science, 104(April 2017), 99–109.

https://doi.org/10.1016/j.ssci.2017.12.018

Gürbüz, F., Özbakir, L., & Yapici, H. (2009). Classification

rule discovery for the aviation incidents resulted in

fatality. Knowledge-Based Systems, 22(8), 622–632.

https://doi.org/10.1016/j.knosys.2009.06.013

Han, J., Kamber, M., & Pei, J. (2011). Data Mining: Concepts

and Techniques. In Data Mining: Concepts and

Techniques. Morgan Kaufmann Publishers Inc.

https://doi.org/10.1016/C2009-0-61819-5

HaSPA (Health and Safety Professionals Alliance). (2012).

The Core Body of Knowledge for the Generalist OHS

Professionals.

Hovden, J., Albrechtsen, E., & Herrera, I. A. (2010). Is there

a need for new theories, models and approaches to

occupational accident prevention? Safety Science, 48(8),

950–956. https://doi.org/10.1016/j.ssci.2009.

06.002

Huan, L., & Hiroshi, M. (2007). Computational Methods of

Feature Selection. Chapman and Hall/CRC Press.

https://doi.org/10.1016/j.ipm.2009.03.003

IATA - IATA Releases 2022 Airline Safety Performance.

(n.d.). Retrieved October 9, 2023, from

https://www.iata.org/en/pressroom/2023-releases/2023-

03-07-01/

ICAO. (1987). Doc 9156-AN/900 Accident / Incident

Reporting Manual.

Jardine, A. K. S., & Tsang, A. H. C. (2013). Maintenance,

Replacement, and Reliability: Theory and Applications.

In Maintenance, Replacement, and Reliability: Theory

and Applications (2nd ed.). CRC Press Taylor and

Francis Group. https://doi.org/10.1201/b14937

Katsakiori, P., Sakellaropoulos, G., & Manatakis, E. (2009).

Towards an evaluation of accident investigation methods

in terms of their alignment with accident causation

models. Safety Science, 47(7), 1007–1015.

https://doi.org/10.1016/j.ssci.2008.11.002

Kauffeld, J. T. L. and K. (2003). Evaluating the Effect of

Technology Insertion into the National Airspace

System.pdf. The Rutgers Scholar, 5.

Larouzee, J., & Le Coze, J. C. (2020). Good and bad reasons:

The Swiss cheese model and its critics. Safety Science,

126(February), 104660. https://doi.org/10.1016/j.ssci.

2020.104660

Leveson, N. (2004). A new accident model for engineering

safer systems. 42, 237–270. https://doi.org/10.

1016/S0925-7535(03)00047-X

Leveson, N. (2012). Engineering a Safer World: System

Thinking Applied to Safety. The MIT Press.

Mathur, P., Khatri, S. K., & Sharma, M. (2017). Prediction of

aviation accidents using logistic regression model.

International Conference on Infocom Technologies and

Unmanned Systems ICTUS: Trends and Future

Directions, 725–728. https://doi.org/10.1109/ ICTUS.

2017.8286102

Maurino, J. T. L. and M. (2001). An aviation system risk

model (ASRM) case study: Air Ontario 1363.

Department of Industrial Engineering, Rutgers

University, Rutgers Scholar, 3, 7. http://rutgersscholar.

rutgers.edu/volume03/maurluxh/maurluxh.htm

Narasimha, M., & Devi, V. . (2011). Pattern recognition: An

Algorithmic Approach. In Undergraduate topics in

computer science. Springer.

Nazeri, Z., Donohue, G., & Sherry, L. (2008). Analysing

Relationships Between Aircraft Accidents and Incidents.

Research in Air Transportation, 1–5.

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.

1.659.111&rep=rep1&type=pdf%5Cnhttp://catsr.ite.gm

u.edu/pubs/NazeriICRAT08_formatted.pdf

Phimster, J. R., Bier, V. M., & Kunreuther, H. C. (2004).

Accident precursor analysis and management: Reducing

technological risk through diligence. THE NATIONAL

ACADEMIES PRESS. https://doi.org/10.

17226/11061

Pyzdek, T. (2017). Six Sigma Handbook A complete guide for

green belts, black belts and managers at all levels

(Fourth Edi). McGraw-Hill.

Qureshi, Z. (2008). A Review of Accident Modelling

Approaches for Complex Critical Sociotechnical

Systems. 12th Australian Workshop on Safety Related

Programmable Systems (SCS’07), Adelaide, 47–59.

Reason, J. (1990). Reason J. - Human Error-Cambridge

University Press (1990) .pdf. Cambridge University

Press 1990.

Reason, J. (2000). Human Error: Models and Management.

BMJ, 320, 768–770.

Revolutionizing Healthcare Industry with Machine Learning.

(n.d.). Retrieved February 28, 2024, from

https://www.datasciencecentral.com/how-machine-

learning-is-revolutionizing-the-healthcare-industry/

Roelen, A. L. C., Lin, P. H., & Hale, A. R. (2011). Accident

models and organisational factors in air transport: The

need for multi-method models. Safety Science, 49(1), 5–

10. https://doi.org/10.1016/j.ssci.2010.01.022

Safety Management System | Federal Aviation

Administration. (n.d.). Retrieved April 10,

Enhancing Aviation Safety Analysis in MROs: A Complex Emergent Model with a Predictive Approach

347

2023, from https://www.faa.gov/about/initiatives/sms/

explained/basis

Shappell, S. A., & Wiegmann, D. A. (2000). The human

factors analysis and classification system (HFACS).

Stoop, J., & Dekker, S. W. A. (2012). Are safety

investigations pro-active? Safety Science, 50(6), 1422–

1430. https://doi.org/10.1016/j.ssci.2011.03.004

Stoop, J. J. a, & Kahan, J. P. J. (2005). Flying is the safest

way to travel: How aviation was a pioneer in independent

accident investigation. Technology, 1944(June), 1–37.

https://doi.org/10.18757/ejtir.

2005.5.2.4392

Underwood, P., & Waterson, P. (2012). A critical review of

the stamp, fram and accimap systemic accident analysis

models. Advances in Human Aspects of Road and Rail

Transportation, January 2016, 385–394. https://doi.

org/10.1201/b12320

Yeun, R., Bates, P., & Murray, P. (2014). Aviation safety

management systems. World Review of Intermodal

Transportation Research, 5(2), 168–196. https://doi.

org/10.1504/WRITR.2014.067234.

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

348