مقاله انگلیسی رایگان در مورد روش های یادگیری ماشین برای پیش بینی حوادث شغلی – الزویر 2018

According to International Labour Organization (ILO) estimation, globally about 2.3 million workers succumb to death annually due to occupational accidents and diseases which include approximately 3.6 lakh fatal accidents [1]. Overall, nearly 337 million occupational accidents are reported per year. From ILO report, it is revealed that approximately 4% of the annual gross domestic product (GDP), which is equivalent to US $1.25 trillion, is drained off due to occupational accidents [2]. From EUROSTAT, it is reported that each year, 3.2% of workers in the European Union, i.e., EU-27 meet an accident at their working places [3]. In relation with this, ILO also makes the following comments: “Fatalities are not fated; accident do not just happen; illness is not random; they are caused” [4]. The basic causes of accidents are unsafe conditions or unsafe acts or both. There are multiple factors contributing towards an accident. There are many theories available in literature that explain the causation of accidents. Khanzode et al. [5] explained the various theories in their study behind the accidents such as accident proneness theory [6], Domino theory [7], injury epidemiology [8], system theory [9], sociotechnical system theory [10], and macro-ergonomic theory [11]. An injury event is occurred due to the presence of a chain of events or causal factors. If the causes are known, the outcomes (i.e., accidents) can be predicted. In addition, the predictive models will quantify the contribution of the various causal factors towards an accident to happen. Predictive models for occupational accidents can be statistical learning based or machine learning (ML) based. Owing to the large amount of data available, ML supersedes traditional statistical counterpart in predicting future events that has been used in various fields such as engineering, medical science, finance, and it renders very useful results [12]. However, a review of literature shows that the ML techniques have been used in occupational accident analysis on a limited basis [13]. So far, studies made on occupational analysis show the use of ML techniques in terms of their predictive power [14] and explanatory capacity [15]. These methods, based on historical data from incident reports, or interview with employees, ensure their advantages over conventional statistics in terms of predictive functions and importance of predictors with a bearing on incident outcomes. The potential benefits of ML can not only be realized from the capability of processing large quantity of data but also from: (i) their capability to deal with large dimensional problems, (ii) their flexibility in reproducing the data generation structure irrespective of complexity, and (iii) their predictive and interpretative potential through the extraction of rules. Due to the capability of ML techniques, it has been used successfully in several domains including occupational accident analyses. However, the ML techniques do not produce good results if their parameters are not tuned.