Accident Narratives Research Articles

Automatically Categorizing Construction Accident Narratives Using the Deep-Learning Model with a Class-Imbalance Treatment Technique

Automatically Categorizing Construction Accident Narratives Using the Deep-Learning Model with a Class-Imbalance Treatment Technique

Railroad accident analysis by machine learning and natural language processing

The evolving complexities of railroad systems also increase their vulnerability to failure from human error. This study compared the outcomes of two workflows that incorporated 11 different machine learning techniques to identify characteristics of railroad operations that are generally associated with human-caused accidents. The first workflow engineered features from the fixed attribute fields of a large railroad accident database and the second applied natural language processing to extract features from the unstructured accident narratives. Both workflows applied a Shapely game-theoretic model to rank the importance of features based on their marginal contribution towards predicting accident cause. Among several interesting findings, some of the most unexpected were that human-caused accidents are generally not associated with high train speeds nor derailment type accidents, and that shoving cars is riskier than pulling cars. Those, and other findings, from this study can inform management decisions, planning, and policies to minimize the risk of human-caused accidents.

Wire Strikes and In-Air Obstacle Collisions During Agricultural Aviation Operations.

INTRODUCTION: Wire strikes and in-air collisions with obstacles are a leading cause of accidents in the aerial application industry. While some of these collisions occur due to unseen obstacles, some pilots report being previously aware of the obstacles that they collide with. Whether or not pilots are aware of obstacles pre-collision is an important factor to inform methods of accident prevention.METHODS: Final reports from the National Transportation Safety Board were analyzed for Part 137 Agricultural Operation accidents that took place between January 2020 and December 2022. A deeper analysis of cases that involved an in-air collision with an obstacle was performed, excluding cases that were attributable to an external cause (e.g., aerodynamic stall). The pilot's awareness of the obstacle pre-accident was inferred from accident narratives if available.RESULTS: Nearly half of all accidents (N = 45 of 107) involved an in-air collision with an obstacle (e.g., wire, tree, pole) as the defining event. In cases where pilot awareness of the obstacle was determinable through the accident report, over half of pilots (N = 21 of 39) had previously seen this obstacle yet still made contact with it.DISCUSSION: In-air obstacle collisions make up a substantial portion of accidents within Part 137 Agricultural Operations. Nearly half of pilots were already aware of the obstacle before collision, indicating that inadequate preparation in scoping the field is not a predominant driver of these events. Instead, these findings suggest that other factors including distractions, high task difficulty, and errors in decision-making may contribute.Baumgartner HM. Wire strikes and in-air obstacle collisions during agricultural aviation operations. Aerosp Med Hum Perform. 2023; 94(11):852-856.

Construction accident prevention: A systematic review of machine learning approaches.

The construction industry is an important productive sector worldwide. However, the industry is also responsible for high numbers of work-related accidents, which highlights the necessity for improving safety management on construction sites. In parallel, technological applications such as machine learning (ML) are used in many productive sectors, including construction, and have proved significant in process optimizations and decision-making. Thus, advanced studies are required to comprehend the best way of using this technology to enhance construction site safety. This research developed a systematic literature review using ten scientific databases to retrieve relevant publications and fill the knowledge gaps regarding ML applications in construction accident prevention. This study examined 73 scientific articles through bibliometric research and descriptive analysis. The results showed the publications timeline and the most recurrent journals, authors, institutions, and countries-regions. In addition, the review discovered information about the developed models, such as the research goals, the ML methods used, and the data features. The research findings revealed that USA and China are the leading countries regarding publications. Also, Support Vector Machine - SVM was the most used ML method. Furthermore, most models used textual data as a source, generally related to inspection reports and accident narratives. The data approach was usually related to facts before an accident (proactive data). The review highlighted improvement proposals for future works and provided insights into the application of ML in construction safety management.

“A Lot Can Go Through Your Mind in a Split-Second”: Survivor Stories of Falling from Height

This paper provides insight into the lived experience of individuals who have survived falls from height. Interpretative phenomenological analysis (IPA) was used to analyze qualitative questionnaire and interview data from four participants and from the author, who is a fellow fall survivor. Overarching themes were “Making sense of the fall”; “Chance and Agency”; “Impact” and “Recover and Reflect.” The analysis identified the importance for survivors of developing a coherent narrative to situate the fall within the life-story. Processes of meaning-making and the role of blame, guilt and fault in the construction of accident narratives were also examined. First-hand accounts of physical and psychological consequences for fall survivors provide medical and therapeutic professionals with an opportunity to improve their understanding of and care for fall survivors.

Deep Learning and Text Mining: Classifying and Extracting Key Information from Construction Accident Narratives

Construction accidents can lead to serious consequences. To reduce the occurrence of such accidents and strengthen the execution capabilities in on-site safety management, managers must analyze accident report texts in depth and extract valuable information from them. However, accident report texts are usually presented in unstructured or semi-structured forms; analyzing these texts manually requires a lot of time and effort, it is difficult to cope with the demand of analyzing a large number of accident texts, and the quality of key information extracted manually may be poor. Therefore, this study proposes a classification method based on natural language processing (NLP) technology. First, we developed a text classification model based on a convolutional neural network (CNN) that can automatically classify accident categories based on accident text features. Next, taking the classified fall accidents as an example, we extracted key information from accident narratives using the term frequency-inverse document frequency (TF-IDF) method and presented it visually using word clouds. The results show that the overall accuracy of the CNN model reaches 84%, which is better than the other three shallow machine-learning models. Then, eight key accident areas and three accident-prone operations were identified using the TF-IDF algorithm. This study can provide important guidance for project managers and can be used for on-site safety management to help prevent production safety accidents.

Contextual Representation in NLP to Improve Success in Accident Classification of Mine Safety Narratives

Contextual representation has taken center stage in Natural Language Processing (NLP) in the recent past. Models such as Bidirectional Encoder Representations from Transformers (BERT) have found tremendous success in the arena. As a first attempt in the mining industry, in the current work, BERT architecture is adapted in developing the MineBERT model to accomplish the classification of accident narratives from the US Mine Safety and Health Administration (MSHA) data set. In the past multi-year research, several machine learning (ML) methods were used by authors to improve classification success rates in nine significant MSHA accident categories. Out of nine, for four major categories (“Type Groups”) and five “narrow groups”, Random Forests (RF) registered 75% and 42% classification success rates, respectively, on average, while keeping the false positives under 5%. Feature-based innovative NLP methods such as accident-specific expert choice vocabulary (ASECV) and similarity score (SS) methods were developed to improve upon the RF success rates. A combination of all these methods (“Stacked” approach) is able to slightly improve success over RF (71%) to 73.28% for the major category “Caught-in”. Homographs in narratives are identified as the major problem that was preventing further success. Their presence was creating ambiguity problems for classification algorithms. Adaptation of BERT effectively solved the problem. When compared to RF, MineBERT implementation improved success rates among major and narrow groups by 13% and 32%, respectively, while keeping the false positives under 1%, which is very significant. However, BERT implementation in the mining industry, which has unique technical aspects and jargon, brought a set of challenges in terms of preparation of data, selection of hyperparameters, and fine-tuning the model to achieve the best performance, which was met in the current research.

A transformers-based approach on industrial disaster consequence identification from accident narratives

Industrial accidents are a major part of the more general technological disasters domain. These events vary from mild to severe and can affect millions of people and cause huge environmental and property damage. New technologies enable novel approaches and applications in the field of disaster management which help prevent, mitigate, and respond to such events. More specifically, transformers-based language models dominate natural language applications with outstanding state of the art performance. In this study an effort is being made to identify and evaluate the ability of transformers to extract and evaluate the consequences and effects of an industrial accident using natural language. The eMARS database is used for training data and a BERT model is used for fine tuning. The results on a multilabel classification task for consequence prediction from accident narratives are promising, achieving an overall weighted F1-score of 0.80.

Railroad accident causal analysis with unstructured narratives using bidirectional encoder representations for transformers

Railroad safety is one critical concern for the railroad industry. Extracting useful information from railroad safety-related textual materials is one significant and essential task. To better understand the contributing factors to the railroad accidents, previous studies have primarily focused on the structured fields, while few of them have developed a thorough analysis of the narratives. In addition, due to the difficulty of understanding the terminologies in the accidents’ narratives, it is challenging to extensively use these narratives as a time-consuming and labor-intensive task. Therefore, this study proposed a novel deep learning approach to consistently leverage the values behind these railroad accident narratives. The proposed method modified the classical Bidirectional Encoder Representations for Transformers (BERT) with the connection of a Deep Neural Network (DNN). To validate the superiority of the proposed BERT-DNN, several additional text classification methods were employed in the real-world railroad accident database. Results demonstrate the proposed method in this study can assign congruous accident causes based on the railroad accidents’ narratives precisely and outperforms previous state-of-the-art text classification approaches. The analytical results, along with proposed methodological framework, can contribute to an in-deep understanding of accident causes for practitioners and academics, and ultimately enhance rail operation safety.

The Importance of Specific Phrases in Automatically Classifying Mine Accident Narratives Using Natural Language Processing

The mining industry is diligent about reporting on safety incidents. However, these reports are not necessarily analyzed holistically to gain deep insights. Previously, it was demonstrated that mine accident narratives at a partner mine site could be automatically classified using natural language processing (NLP)-based random forest (RF) models developed, using narratives from the United States Mine Safety and Health Administration (MSHA) database. Classification of narratives is important from a holistic perspective as it affects safety intervention strategies. This paper continued the work to improve the RF classification performance in the category “caught in”. In this context, three approaches were presented in the paper. At first, two new methods were developed, named, the similarity score (SS) method and the accident-specific expert choice vocabulary (ASECV) method. The SS method focused on words or phrases that occurred most frequently, while the ASECV, a heuristic approach, focused on a narrow set of phrases. The two methods were tested with a series of experiments (iterations) on the MSHA narratives of accident category “caught in”. The SS method was not very successful due to its high false positive rates. The ASECV method, on the other hand, had low false positive rates. As a third approach (the “stacking” method), when a highly successful incidence (iteration) from ASECV method was applied in combination with the previously developed RF model (by stacking), the overall predictability of the combined model improved from 71% to 73.28%. Thus, the research showed that some phrases are key to describing particular (“caught in” in this case) types of accidents.

Characterizing accident narratives with word embeddings: Improving accuracy, richness, and generalizability

Introduction: Ensuring occupational health and safety is an enormous concern for organizations, as accidents not only harm workers but also result in financial losses. Analysis of accident data has the potential to reveal insights that may improve capabilities to mitigate future accidents. However, because accident data are often transcribed textually, analyzing these narratives proves difficult. This study contributes to a recent stream of literature utilizing machine learning to automatically label accident narratives, converting them into more easily analyzable fields. Method: First, a large dataset of accident narratives in which workers were injured is collected from the U.S. Occupational Safety and Health Administration (OSHA). Word embeddings-based text mining is implemented; compared to past works, this methodology offers excellent performance. Second, to improve the richness of analyses, each record is assessed across five dimensions. The machine learning models provide classifications of body part(s) injured, the source of the injury, the type of event causing the injury, whether a hospitalization occurred, and whether an amputation occurred. Finally, demonstrating generalizability, the trained models are deployed to analyze two additional datasets of accident narratives in the construction industry and the mining and metals industry (transfer learning). Practical Applications: These contributions improve organizations’ capacities to rapidly analyze textual accident narratives.

A C-BiLSTM Approach to Classify Construction Accident Reports

The construction sector is widely recognized as having the most hazardous working environment among the various business sectors, and many research studies have focused on injury prevention strategies for use on construction sites. The risk-based theory emphasizes the analysis of accident causes extracted from accident reports to understand, predict, and prevent the occurrence of construction accidents. The first step in the analysis is to classify the incidents from a massive number of reports into different cause categories, a task which is usually performed on a manual basis by domain experts. The research described in this paper proposes a convolutional bidirectional long short-term memory (C-BiLSTM)-based method to automatically classify construction accident reports. The proposed approach was applied on a dataset of construction accident narratives obtained from the Occupational Safety and Health Administration website, and the results indicate that this model performs better than some of the classic machine learning models commonly used in classification tasks, including support vector machine (SVM), naïve Bayes (NB), and logistic regression (LR). The results of this study can help safety managers to develop risk management strategies.

Text mining-based construction site accident classification using hybrid supervised machine learning

Safety is one key consideration in the monitoring of construction projects by engineers. Accidents in the project can potentially cause issues, such as workers' injury and progress delay, which lead to financial losses. Generally, accident narratives store all summaries and causes of the related events. Since documentations rapidly use large quantities of resources, the implementation of Artificial Intelligence (AI) begins to seek attention. Nevertheless, in current models, there are still drawbacks, such as weak learning performance and substantial error rate. In this regard, this study develops a hybrid model incorporating Gated Recurrent Unit (GRU) and Symbiotic Organisms Search (SOS), named Symbiotic Gated Recurrent Unit (SGRU). SOS searches the best parameters of GRU to ensure optimal performance. Furthermore, Natural Language Processing is applied to pre-process the text data prior classification process. The experimental result in this study showcases SGRU as the best classification model among other AI models. Therefore, SGRU shares the capability to aid the safety assessments of construction projects.

Electrically Hazardous Jobs in the U.S.

This article examined 1004 U.S. Occupational Safety and Health Administration cases of electrical injury that occurred between January 1, 2011 and June 28, 2017. The data are examined to determine which occupations experienced fatal electrical injury. Characteristics such as the degree of injury, the nature of injury, and task assignments at the time of injury were determined. There were 705 fatal electrical injuries from all causes. There were 17 occupations with ten or more fatal cases, led by, in order, “electricians,” “construction laborers,” “laborers, except construction,” “electrical power installers and repairers,” and “tree trimming occupations.” In addition, the data are also examined to isolate fatal overhead power line injuries. Overhead power lines were involved in 246 fatal electrical shock injuries to workers and ten burn injuries. Occupations most affected by overhead power line injuries were studied in greater detail. Just six occupations account for more than one half of all those fatally injured by electricity in overhead power line cases. Occupations with ten or more fatal overhead power line injuries include, in order, “construction laborer,” “laborers, except construction,” “tree trimming occupations,” “electrical power installers and repairers,” “roofers,” and “truck drivers, heavy.” Several identified occupations with unexpected electrical exposure, such as “roofers,” “tree trimmers,” and “truck drivers,” are not usually associated with electrically hazardous jobs. Erecting, moving, and dismantling scaffolding, the use of conductive ladders, booms, and tag lines, and tree trimming in the vicinity of overhead power lines were common scenarios in which the fatal electrical injuries are studied. Working without deenergizing circuits, working without proper prejob hazard analysis and planning, and working without proper personal protective equipment also were mentioned in the accident narratives.

Deep learning and network analysis: Classifying and visualizing accident narratives in construction

If headway is to be made to improve safety performance in construction, then there is a need to learn from past accidents. Accident reports provide a useful source of information to make sense as to why and how events occurred. Analyzing such reports, however, can be a lengthy and challenging process as there is a tendency for data to be presented in an unstructured or semi-structured free-text format. Thus, being able to classify and analyze the narrative that surrounds accidents and to better understand their causal nature is a challenge. Text classification using shallow machine learning with sophisticated manual lexical, syntactic, and semantic features engineering has been typically used to mine accident data. However, this approach requires highly skilled experts with domain knowledge to undertake this task. A limited number of studies have employed deep learning models to examine the text of safety reports in construction. In consideration of this limitation, word embedding is used to model the semantic narratives of accidents. Then, a Convolution Neural Network (CNN) model is trained to automatically extract text features and classify accident narratives without manual feature processing. The Latent Dirichlet Allocation (LDA) model is used to examine the interdependency that exists between causal variables to visualize the accident narratives. The proposed automated classification model and LDA-based network analysis method provide a useful approach to enable machine-assisted interpretation of texts-based accident narratives. Moreover, the proposed approach can provide managers with much-needed information and knowledge to improve safety on-site.

Intelligent human-machine approaches for assigning groups of injury codes to accident narratives

In injury surveillance, different aspects of an injury event are captured using injury codes such as the External-cause-of-injury (E-code), Major Injury Factor (MIF), and Intent. These are usually assigned by human coders based on accident narratives. Previous studies have examined automated and semi-automated filtering approaches that use machine learning (ML) models to assign single E-codes to accident narratives. In this study, our goal was to examine the effectiveness of these approaches for assigning groups of injury codes. This was done for three different types of injury codes (E-code, MIF, and Intent) using several ML models (Logistic Regression, Support Vector Machine, and Long-Short-Term-Memory based Recurrent-Neural-Network). Four filtering strategies were also tested which used the probability of prediction correctness assigned by the Logistic Regression model. These approaches were evaluated for a manually-coded dataset, provided by the Queensland Injury Surveillance Unit containing about half a million injury cases. The results showed very similar performance for the three ML models. The overall sensitivity of each model was quite high and almost identical for E-code (0.81–0.82), MIF (0.69–0.71), and Intent (0.96–0.97). However, the unweighted sensitivities were lower - E-code (0.67–0.75), MIF (0.59–0.62), and Intent (0.46–0.56), reflecting a general trend of each model to under-predict small categories. It was also observed that the probability of correctly assigning all three codes was low (0.58). Filtering approaches resulted in large improvements in sensitivity for smaller categories and the probability of predicting all three codes correctly.

Construction accident narrative classification: An evaluation of text mining techniques

Learning from past accidents is fundamental to accident prevention. Thus, accident and near miss reporting are encouraged by organizations and regulators. However, for organizations managing large safety databases, the time taken to accurately classify accident and near miss narratives will be very significant. This study aims to evaluate the utility of various text mining classification techniques in classifying 1000 publicly available construction accident narratives obtained from the US OSHA website. The study evaluated six machine learning algorithms, including support vector machine (SVM), linear regression (LR), random forest (RF), k-nearest neighbor (KNN), decision tree (DT) and Naive Bayes (NB), and found that SVM produced the best performance in classifying the test set of 251 cases. Further experimentation with tokenization of the processed text and non-linear SVM were also conducted. In addition, a grid search was conducted on the hyperparameters of the SVM models. It was found that the best performing classifiers were linear SVM with unigram tokenization and radial basis function (RBF) SVM with uni-gram tokenization. In view of its relative simplicity, the linear SVM is recommended. Across the 11 labels of accident causes or types, the precision of the linear SVM ranged from 0.5 to 1, recall ranged from 0.36 to 0.9 and F1 score was between 0.45 and 0.92. The reasons for misclassification were discussed and suggestions on ways to improve the performance were provided.

Industrial Safety and Utopia: Insights from the Fukushima Daiichi Accident.

Feedback from industrial accidents is provided by various state or even international, institutions, and lessons learned can be controversial. However, there has been little research into organizational learning at the international level. This article helps to fill the gap through an in-depth review of official reports of the Fukushima Daiichi accident published shortly after the event. We present a new method to analyze the arguments contained in these voluminous documents. Taking an intertextual perspective, the method focuses on the accident narratives, their rationale, and links between "facts," "causes," and "recommendations." The aim is to evaluate how the findings of the various reports are consistent with (or contradict) "institutionalized knowledge," and identify the social representations that underpin them. We find that although the scientific controversy surrounding the results of the various inquiries reflects different ethical perspectives, they are integrated into the same utopian ideal. The involvement of multiple actors in this controversy raises questions about the public construction of epistemic authority, and we highlight the special status given to the International Atomic Energy Agency in this regard.

Text Mining the Contributors to Rail Accidents

Rail accidents represent an important safety concern for the transportation industry in many countries. In the 11 years from 2001 to 2012, the U.S. had more than 40 000 rail accidents that cost more than $45 million. While most of the accidents during this period had very little cost, about 5200 had damages in excess of $141 500. To better understand the contributors to these extreme accidents, the Federal Railroad Administration has required the railroads involved in accidents to submit reports that contain both fixed field entries and narratives that describe the characteristics of the accident. While a number of studies have looked at the fixed fields, none have done an extensive analysis of the narratives. This paper describes the use of text mining with a combination of techniques to automatically discover accident characteristics that can inform a better understanding of the contributors to the accidents. The study evaluates the efficacy of text mining of accident narratives by assessing predictive performance for the costs of extreme accidents. The results show that predictive accuracy for accident costs significantly improves through the use of features found by text mining and predictive accuracy further improves through the use of modern ensemble methods. Importantly, this study also shows through case examples how the findings from text mining of the narratives can improve understanding of the contributors to rail accidents in ways not possible through only fixed field analysis of the accident reports.

The Deepwater Horizon explosion: non-technical skills, safety culture, and system complexity

The explosion and destruction of the Deepwater Horizon (DH) was a watershed moment for safety management in the US oil and gas industry. The 2011 National Oil Spill Commission investigation identified a range of operational behaviours and underlying safety management problems that were causal to the mishap. Yet, to date these have not been systematically considered within a human factors framework. To achieve this, we draw upon two applied psychology domains that are highly influential within safety research. First, we apply non-technical skills (NTS) (social and cognitive skills that underpin safe performance in complex work environments) theorem to understand operational activities in the lead-up and occurrence of the well blowout. NTS research is used to develop interventions for training and observing safety behaviours (e.g. decision-making, teamwork). Second, we apply safety culture theory to understand how the organisational and industry environment shaped the management of risk. Safety culture research is used to understand and change the socio-technical constraints and enablers of safety activity in high-risk workplaces. Finally, to integrate these perspectives, we take a systems-thinking perspective to understand the mishap. A common critique of accident narratives is their failure to systematically consider how the components of an incident interact together to escalate risk. From a systems-thinking perspective, understanding the interactions leading to the DH mishap is crucial for ensuring interventions are effective in preventing future mishaps. We develop an accident model that captures the various interactions and system factors leading to the blowout.