Machine learning methods can establish complex nonlinear relationships between input and response variables for stadium fire risk assessment. However, the output of machine learning models is considered very difficult due to their complex “black box” structure, which hinders their application in stadium fire risk assessment. The SHapley Additive exPlanations (SHAP) method makes a local approximation to the predictions of any regression or classification model so as to be faithful and interpretable, and assigns significant values (SHAP value) to each input variable for a given prediction. In this study, we designed an indicator attribute threshold interval to classify and quantify different fire risk category data, and then used a random forest model combined with SHAP strategy in order to establish a stadium fire risk assessment model. The main objective is to analyze the impact analysis of each risk characteristic on four different risk assessment models, so as to find the complex nonlinear relationship between risk characteristics and stadium fire risk. This helps managers to be able to make appropriate fire safety management and smart decisions before an incident occurs and in a targeted manner to reduce the incidence of fires. The experimental results show that the established interpretable random forest model provides 83% accuracy, 86% precision, and 85% recall for the stadium fire risk test dataset. The study also shows that the low level of data makes it difficult to identify the range of decision boundaries for Critical mode and Hazardous mode.

Stadium fires can easily cause massive casualties and property damage. The early risk prediction of stadiums will be able to reduce the incidence of fires by making corresponding fire safety management and decision making in an early and targeted manner. In the field of building fires, some studies apply data mining techniques and machine learning algorithms to the collected risk hazard data for fire risk prediction. However, most of these studies use all attributes in the dataset, which may degrade the performance of predictive models due to data redundancy. Furthermore, machine learning algorithms are numerous and applied to fewer stadium fires, and it is crucial to explore models suitable for predicting stadium fire risk. The purpose of this study was to identify salient features to build a model for predicting stadium fire risk predictions. In this study, we designed an index attribute threshold interval to classify and quantify different fire risk data. We then used Gradient Boosting-Recursive Feature Elimination (GB-RFE) and Pearson correlation analysis to perform efficient feature selection on risk feature attributes to find the most informative salient feature subsets. Two cross-validation strategies were employed to address the dataset imbalance problem. Using the smart stadium fire risk data set provided by the Wuhan Emergency Rescue Detachment, the optimal prediction model was obtained based on the identified significant features and six machine learning methods of 12 combination forms, and full features were input as an experimental comparison study. Five performance evaluation metrics were used to evaluate and compare the combined models. Results show that the best performing model had an F1 score of 81.9% and an accuracy of 93.2%. Meanwhile, by introducing a precision-recall curve to explain the actual classification performance of each model, AdaBoost achieves the highest Auprc score (0.78), followed by SVM (0.77), which reveals more stable performance under such imbalanced data.

Although the interest in metallurgical accident investigation of blast furnace gas (BFG) leakage has increased to explore the engineering failures, more effort is needed to address the individual and organizational causative factors to clear and determine the weak links for improving safety management and accident prevention to achieve green metallurgical manufacturing. This study aims to examine the causative factors and evolution paths of BFG leakage by introducing a combined method, the 24 model and Bayesian network (BN), based on 50 cases of fire, explosion and suffocation accidents caused by BFG leakage. A BN model of BFG leakage was established based on the identification of 25 causative factors by the 24 model. Results showed that eight nodes, including A1 (unsafe operation), A2 (unsafe behavior), A4 (unsafe condition), B1 (valve failure), B2 (improper gas safety operation), X4 (use of BFG violates regulations), X5 (water gas is not cut off before shutdown reduction) and X6 (incomplete steam purging), were more sensitive than others, and the posterior probability of nodes A1, A2, A3 (unsafe command), A4, B1, B2, B4 (improper emergency behavior), B5 (unsafe behaviors on BFG site) increased compared to prior probability. Three main accident causal chains were obtained which indicate that control the unsafe operations (A1) related to gas (B2) and valve (B1) are suggested to be improved. Another important factor is A4 (unsafe condition), which is related to intrinsic safety conditions. Considering the results, the key points of 3E strategy about BFG leakage prevention are suggested. This study provides useful insights to understand the organizational and individual factors and their relative influence in BFG leakage accidents, which will support BFG leakage prevention and safety management.

Introduction: The digital inclusion of middle-aged and older adults, while socially beneficial, exposes this population with relatively lower digital literacy to increased cybersecurity risks. However, the current understanding of the psychosocial mechanisms that undermine their digital risk perception (DRP) remains fragmented. Methods: Grounded in the Interaction of Person-Affect-Cognition-Execution (I-PACE) model, this study explored the interrelations among key psychological and behavioral factors associated with DRP. Data from 356 Chinese middle-aged and older adults aged 45 years and above in Chongqing were analyzed using structural equation modeling (SEM) combined with regression-based SHAP analysis. Results: Mobile phone dependence (MPD) was negatively associated with DRP (b = -0.253, p < 0.001). Loneliness and low perceived social support were correlated with greater fear of missing out (FoMO) and higher MPD, which were in turn linked to lower DRP. The findings also revealed a correlational pattern suggesting bidirectional tendencies between MPD and FoMO. The SHAP interpretation indicated that loneliness was the strongest correlate of DRP, followed by MPD and FoMO. Discussion: The identified pattern aligns with the process logic of the I-PACE framework. Overall, the study delineates an integrated pattern of psychological and behavioral factors influencing DRP and provides a conceptual basis for strengthening digital safety awareness and literacy among middle-aged and older adults.

In order to solve the problem of the low action recognition accuracy of passengers’ unsafe behaviors caused by redundant joints, this study proposes an efficient recognition method based on a Kinect sensor. The method uses the pelvis as the starting point of the vector and high-frequency bone joints as the end point to construct the recognition feature vector. The joint angle difference between actions is obtained by using the cosine law, and the initial test result is converted into action similarity combined with the DTW similarity algorithm. Considering the combination of 3 angle features and 4 joint feature selection methods, 12 combined recognition models are formed. A comparative experiment was carried out to identify five types of unsafe behaviors of metro passengers—punch, one-armed call for help, arms for help, trip forward and trip backwards. The results show that the overall selection of joints has a poor similarity effect and cannot achieve the purpose of recognition. The overall recognition model effect of the local “pelvis divergence method” is higher than that of the local “adjacent joint method”, and the local “pelvis divergence method” has the highest recognition result of the maximum angle difference model, and the recognition results of the five unsafe behaviors are 86.9%, 89.2%, 85.5%, 86.7%, and 88.3%, respectively, and the recognition accuracy of this method is 95.7%, indicating the feasibility of the model. The recognition results are more concentrated and more stable, which significantly improves the recognition rate of metro passengers’ unsafe behavior.

This study aims to examine the risk of each category of behavior related to different accident types and consequences by introducing the Monte Carlo algorithm. Totally about 234 cases were collected to construct a structured dataset, and 12 categories of passengers’ unsafe behaviors related to the four types of metro operation accidents including escalator injury, crowd stampede, pinched by a shielding barrier and fall injury were considered for this analysis. Results showed that although the total behavioral risk of each accident type obeys normal distribution with means between (4, 6] in the Moderate risk level, the variances of pinched by a shielding barrier (1.29) and crowded stampede (1.08) are much higher than escalator injury (0.60) and fall injury (0.59). The top four contributors that resulted in the accident behavioral risks were irregular behavior on the escalator, go on boarding even after the door-close alarm sounded, push each other, and pay no attention to the ground. It was concluded that due to the uncertainty of the behavior occurrence and its consequences, there are differences of the probabilities in different risk levels of the four accidents caused by the various passengers’ unsafe behaviors. Based on the results, risk mitigation measures for different accident types are recommended. This paper provides a quantitative method to compare and analyze the effect of passengers’ unsafe behaviors related to different accident types and severities, which has relative comparative significance and improves the accuracy of judging the probability distribution of risk degree that can provide basis for risk mitigation in metro operation.