Risk Assessment and Machine Learning Models

Abstract

Safety management system for operational domains with extensive scope is a challenging issue. In the current safety management system literature, there are not efficient methods to implement subjective inferences of organization experts to analyze data and predict the performance of safety and quality management systems by using all available data and identify weakness points to improve organizational safety and quality process. Machine learning is a subset of artificial intelligence that involves the algorithms and model development to allow safety and quality management systems that improve their performance over time automatically. The benefits of machine learning include efficiency improvement, reduced costs, improved decision-making, and increased innovation to study the future of risk management. The existence of massive data in operational sectors is a critical challenge for machine learning implementation in safety risk management systems. A safety data pool from all occurrences and hazards should be designed to improve and know about the future of safety management system, and then design the system to assess, analyze, verify, and predict the safety assessment result to decide proper management system decisions. In this study, the machine learning method implementations are proposed for risk management in operational domains.

1. Introduction

Serious incidents or accidents often occur during flight operations, which have severe consequences. The International Air Transport Association (IATA) has developed standards for analyzing safety data to monitor and emphasize the scope of flight safety concerns. There are some areas that benefit from this study in reviewing and analyzing the safety index, such as authorities and insurance companies.

ICAO itself recognized the framework for the safety management system (SMS), Figure 1 show the components and elements of SMS framework.

1. Safety policy and objectives.

2. Safety risk management.

3. Safety assurance.

4. Safety promotion.

Figure 1.

Safety management system component.

Safety is clarified as “the state in which the possibility of harm to persons or of property damage is reduced to, and maintained at or below, an acceptable level through a continuing process of hazard identification and safety risk management.” SMS implementation starts with hazard identification. Hazards and their related outcome are essential to the implementation of safety and quality risk management [1]. A safety risk is measured by the likelihood and severity appointed to the safety consequence. Safety risk mitigations are proposed to decrease the level of safety risk. There are a few possible risks that can occur during a production process [2].

The implementation of the safety risk management process needs to identify hazards. Obvious hazard identification and their related outcome are necessary to implement safety risk management. A hazard is generically identified by safety experts as a state or a statute with the potential to bring death, injuries to personnel, harm to equipment or structures, loss of material, or reduction of the ability to carry out activities. Hazards are available at all stages in the organization and will be identified or detected by using reporting systems, inspections, or audits. As a result, hazards should be detected before they cause accidents, incidents, or other safety-related occurrences. A main method to identify proactive hazards is a voluntary hazard/incident reporting system.

Safety risk management includes the evaluation and mitigation of safety risks. The goals of safety risk management are risk assessment, which is related to hazard identification, and effective and suitable mitigations that need to be implemented. The key component of the safety management process is safety risk management at both the regulatory body and product/service provider level.

In recent years, machine learning algorithms have solved domain-specific problems in various fields. Analyze and predict safety risk assessment in the aviation industry by machine learning algorithms usage is increased and also used.

For decision-making, business understanding, data understanding, data preparation, modeling, evaluation, and deployment are the methods to include machine learning [3]. Risk assessment usage is applied in many application scopes, and many frameworks, methods, and particular applications have been studied in the scientific literature [4].

One of the most important methods to provide significant benefits for businesses and individuals is machine learning, which has been a rapidly growing field in recent years. From improved efficiency to increased innovation, machine learning has the potential to revolutionize the way to increased innovation. Lack of understanding and expertise, cost of implementation, and ethical concerns when adopting machine learning are challenges and issues to use machine learning; despite these challenges, the potential benefits of machine learning are too great. This chapter presents the importance of machine learning in implementing a safety management system used to predict the safety risk trend in the aviation industry. It also initiates the appeal that is utilized by the authors of this book to arrange the chosen literature into different classifications.

2. Safety risk assessment model

Safety risk assessment is a key point in safety-related industries. However, it encounters a number of concerns, partially associated with technological progress and necessary increment. There is currently a demand to assess risk, improve past lessons learning, and identify methods to analyze relevant data. Data are collected with sufficient ability to converse with unexpected events and supply suitable support to empower safety risk management [5].

The likelihood and severity of the consequence or outcome from an identified hazard are measured in safety risk assessment. It is necessary to provide and propose risk mitigation for the identification of such layered consequences.

2.1 Safety risk probability

“The likelihood or frequency that a safety consequence or outcome might occur” is defined as safety risk probability. The probability assessment is the first step in the process of controlling safety risks.

2.2 Safety risk severity

“The harmness which is happened as a consequence or outcome of the detected hazard” is defined as safety risk severity. Once the probability assessment has been completed, the safety risk severity assessment is the next step is to consider the potential consequences related to the hazard.

2.3 Safety risk tolerability

The safety risk probability and severity assessment process can be applied to extract a safety risk index. The index made using the method identified above consists of an alphanumeric designator, indicating the results of the probability and severity analysis.

A safety risk assessment matrix is used to obtain a risk index and then exported to a safety risk tolerability matrix. Safety risk management comprises the assessment and risk analysis to suggest and implement mitigation of safety risks.

The assessment of the risks is the main target of safety risk management at both the state and product/service provider levels, and it is associated with identified hazards to develop and implement efficient and effective mitigations.

2.4 Safety management system

A systematic attitude to manage safety, including the necessary organizational structures, accountabilities, policies, and procedures.

2.5 Risk mitigation

The procedure of compounding defenses or preventive controls to lower the severity and/or likelihood of a hazard’s projected outcome/consequence.

2.6 Human factors (HF)

The main and specific part of the safety analysis process are SSPs and SMSs with human and organizational factors, which are effective risk management systems. Human factors (HF) is an essential point to assure that existing or recommended defenses have been effective to decrease the safety risk. Where necessary, a supplementary HF analysis may be conducted to support that particular risk mitigation exercise/team. Analysis and error definition and error categorization use the human error model to allow the root cause of all hazards and better knowing the result. This conception ensures the sufficient fulfillment of a root cause analysis. The result of human factors analysis can help the safety expert identify root causes and make better decisions for top managers. A more comprehensive and in-depth mitigation process is the important human factors perspective results.

3. Machine learning implementation

The undertaking applications and prompt development in many scope of machine learning (ML) are the reasons which is received a lot of attention recently.

Machine learning (ML) is a method that can learn a task without being explicitly programmed. This makes it very strong and attractive. ML algorithms spontaneously learn an activity from data. They perform some form of inductive inference whereby, after being trained on a dataset, they make predictions for inputs outside the examples observed in the dataset. ML can achieve feasible and cost-effective tasks that could be cost-prohibitive to solve with explicit systems [6].

Machine learning (ML) has received a lot of attention recently due to its agile progress and promising applications in many scopes [7].

Data-driven analysis is an essential tool for both operational efficiency and improved safety and quality practices. Growth in the industry, which is connected with technology progress and data science techniques, has led the operational industry to a more data-driven approach. Safety and quality experts should access to many sources of safety and quality-related information to assess and analyze—significant methods currently available leverage large data sets to identify anomalous business behavior. In recent years, data mining and machine learning methods have been implemented to analyze safety, incident and accident investigation, and error detection in the community. There are many techniques in machine learning (Figure 2).

Figure 2.

Data mining methods.

Figure 2 shows that there are two techniques in data mining models: conventional and data-driven techniques for smart grid analysis. Each method has a specific benefit according to the application scope.

Safety and quality data are collected and processed from many resources such as operational safety and quality assessment and assurance. Safety and quality reports reside in both the specific Safety Action Program and the National Systems coordinated Safety Reporting Systems initiative. This process contains data preprocessing, a high-frequency event evaluation, feature vector generation, algorithm implementation, and post-processing of the results [8].

Machine learning (ML) recognize to be one of the only technically and economically practical resolving to robotize some complicated activities usually discover by humans, such as driving vehicles and recognizing voice. However, these methods face new potential risks and have only been accomplished in systems where the benefits of the methods are known to be worth this increase of risk [9]. Some researchers categorize ML safety into three plans: (1) intrinsically safe methods, (2) method performance and strength, and incorporate (3) run-time error detection methodology [7].

Machine learning as a methodology can detect criteria and schemes in data, predict future data, and help experts make better decisions under uncertainty. Machine learning use statistics sciences and computers to evaluate complicated functions and a decreased reliance on proving confidence intervals around these activities [10]. Machine learning is applied as suitable tool to achieve human-like choice or even super-human performance with automation on specific activities. Data-driven approach is currently have enormous challenge in all industry scope. Enormous growth show that machine learning (ML) has many advancements in computer vision and other fields. Training a neural network is done by using very big datasets to implement machine learning [11].

Currently, a variety of machine learning methods can be used to implement SMS phases including: risk identification, risk analysis, and risk evaluation. Machine learning use data (such as historical and real-time data) for providing inputs to traditional risk assessment techniques. Machine learning is used and applied in many industries such as automotive, aviation, construction, and railways. Paltrinieri et al. [5] studied the knowledge gap by performing a structured review of relevant literature on machine learning usage for risk assessment. The results show that 11 journal papers were published in the Journal of Accident Analysis and Prevention, which makes it the most contributing journal reference. The United States of America, Chinese, and South Korean institutes were the highest level of following affiliations. The adoption of machine learning to implement risk assessment is studied in the automotive industry (over 20% of published articles). The most popular machine learning algorithm selected to accomplish risk assessment was artificial neural networks (ANNs) which are used by support vector machines (SVMs). Historical datasets are practiced in more than 70% of papers, and real-time data to construct and design the machine learning model are implemented in more than 20% of paper. A case-study approach to machine learning model implementation is used in about half of the proposed methods, and about one-fourth have applied their models in a real-world setting.

The results of this review show that the machine learning techniques usage to assess the future of risk is a new approach to study and has an increasing trend in academic publications. Classic risk assessment is supported by machine learning methods and data-driven preparation as inputs. In the future, the operational domains in the industry will need to assess risk in a real-time manner with machine learning methods adoption. The Safety regulatory bodies can use machine learning in risk assessment to validate procedures [5]. Data mining and its key enabler, machine learning, and techniques are the main and usable tools for predicting because these algorithms effectively identify and capture complex patterns and relationships. However, they can account for variable-related assumptions and other limitations, such as a lack of ability to explain things and transparency. Some factors can be present in the data such as environmental factors, human performance, and instrument.

To predict risk categories by using safety and quality data, machine learning algorithms were implemented to classify records in the data set. The prediction of risk categories is determined by using multiple models to be implemented and tested.

To start, features were selected for inclusion in the model. Once features were selected, data was restructured into a valid template for each model’s feature. Each model applied by machine learning shall be trained, and tested, and has its own strengths and weaknesses. Overall, when selecting a suitable model, it is the main issue to propound the strengths and weaknesses of each method and arrangement to ensure poorly fit or heavily biased models are not used.

In conclusion, all the models that were trained demonstrate some degree of reliability in accurately predicting risk [12].

4. Conclusions

The main contribution of this chapter is considering all the characteristics that impact operational performance in a systematic evaluation of safety risk assessment processes while using the machine learning multiclass classification models to predict several levels of classes related to predicting the safety risk assessment over time.

The results of this chapter study also suggest that the application of the machine learning models such as deep learning model and multiclass classification models to safety and quality data not only offers improved data classification but also provides a framework for better safety and quality management systems implementation. Operational companies can employ this model to predict safety risk performance before the operation in order to prevent or decrease any risks of hazards.