دانلود رایگان مقاله انگلیسی کاربرد های استفاده از روش های یادگیری ماشین برای پیش بینی حوادث شغلی - الزویر 2018

Abstract

Although, the usefulness of the machine learning (ML) technique in predicting future outcomes has been established in different domains of applications (e.g., heath care), its exploration in predicting accidents in occupational safety domain is almost new. This necessitates the investigation of ML techniques in predicting accidents. But, ML-based algorithms cannot produce best performance until its parameters are properly tuned or optimized. Moreover, only the selection of efficient optimized classifier may not fulfil the overall decision-making purposes as it cannot explain the interrelationships among the factors behind the occurrence of accidents. Hence, in addition to prediction, decision making rules are required to be extracted from the accident data. Considering the abovementioned issues, in this research, optimized machine learning algorithms have been applied to predict the accident outcomes such as injury, near miss, and property damage using occupational accident data. Two popular machine learning algorithms, namely support vector machine (SVM) and artificial neural network (ANN) have been used whose parameters are optimized by two powerful optimization algorithms, namely genetic algorithm (GA) and particle swarm optimization (PSO) in order to achieve higher degree of accuracy and robustness. PSO-based SVM outperforms the other algorithms with highest level of accuracy and robustness. Furthermore, rules are extracted by incorporating decision tree C5.0 algorithm with PSO-based SVM model. Finally, a set of nine useful rules extracted to identify the root causes behind the injury, near miss and property damage cases. A case study from a steel plant is presented to reveal the potentiality and validity of the proposed methodology.

6. Conclusion and future scope

In this present study, optimized machine learning-based prediction models have been developed to predict the incident outcomes at workplace. Two powerful and effective classifiers, namely SVM and ANN have been used for this task whose parameters are optimized by two popular optimizers, namely GA and PSO. The findings of this research work put forward some useful insights on data preprocessing tasks, parameter optimizations of classifiers, and rule extraction from the accident data. For examples, findings of the analysis reveal that PSO-based SVM outperforms other classifiers in terms of accuracy (i.e., accuracy of PSO-based SVM is 90.67%). In addition, using sensitivity analysis, PSO-SVM is found to be the most robust classifier as well. Furthermore, rules obtained from the PSO-SVM based C5.0 are also found to be effective as they can be used for the interpretation of the factors in terms of rules behind the incident occurrences. Some key findings from the analysis show that slipping is the common cause for injury cases (as observed from Topic 9). Other than slipping, other causes including collision, electric flash, and road incidents are identified for the injuries in some of the divisions of the steel plant. Slipping issues remain also the primary cause for the near miss cases for some of the divisions. In addition, some days of the week like Saturday, Monday, and Tuesday are also identified where near miss incidents happen more frequently. In Div2, Div9, Div10, and Div12, crane operation, falling from height, fire incidents, pipe leakage and vehicle collision are identified that cause to the occurrence of near miss cases in plant. Moreover, collision and electrical flash are the primary causes that leads to property damage more in some of the divisions- Div3 and Div8.