**4. Result and Discussion**

Based on the setting latent variables and observed variables (K1–K6, S1–S7, and M1–M6), r1~r4 was set as the residual variable of the corresponding latent variable, and e1~e23 was the residual variable of the corresponding observed variable. The model was tested by IBM SPSS AMOS21.0 software (International Business Machines Corporation, Armonk, NY, USA). First, the evaluation data of the structural equation model were calculated, and the model was modified by adjusting the model to meet all kinds of fitting indexes. This paper mainly used the absolute fit index, the comparative fit index, and the parsimonious fit index to judge the fitness of the model [40]. After several revisions of the model and calculation of the fitting data, the final fitting results were obtained. The model fitting reference criteria and the final model fitting data are shown in Table 5.


**Table 5.** Model fitting criteria and results.

Notes: RMSEA: Root-Mean-Square Error of Approximation; NFI: Comparative Fit Index; RFI: Relative Fit Index; IFI: Increasing Fit Index; TLI: Tucker-Lewis Index; CFI: Comparative Fit Index; PGFI: Goodness of Fit Index.

The result of the model modification shows that the fitting data of the model meets the requirements of the fitting criteria, which shows that the model has good fit [41]. Based on the above analysis, the result of the final model is shown in Figure 4.

**Figure 4.** Final evaluation model of targeted intervention nodes for unsafe behavior.

According to the final output results of the evaluation model, the path coefficients of H1–H6 were 0.766, 0.81, 0.77, 0.67, 0.48, and 0.59, respectively, which proved that H1–H6 were significant influences. Based on the path coefficients of H1–H3, the order of the effects of the three target intervention nodes was as follows: single-dimensional feature nodes > multi-dimensional feature nodes > key work behavior nodes.

The results of H1 showed that the positive influence of key work behavior on the prediction of intervention effect was supported, and the accuracy and practicability of targeted intervention nodes K1–K5 were verified. The results of Judi et al. show that the precise positioning of specific unsafe behaviors and the positive strengthening of specific safety practices can effectively reduce

occupational accidents [42]. This conclusion is in line with the research results of this paper. There is a very complicated nonlinear relationship between the overall safety state of coal mining enterprises and various unsafe behaviors of miners [43]. Intervening unsafe behavior that can easily lead to coal mine accidents has a positive effect on the overall safety state of coal mines. The key work behaviors K1–K5 involved in H1 come from different groups of workers, including coal mine workers, field commanders, and managemen<sup>t</sup> staff, all of which have a positive support effect on the prediction results. The results support Luria's view that enhancing the visibility of workers' behavior helps to increase the impact of intervention programs [44]. It is obvious that locating the key types of behavior intervention nodes is an effective way to enhance the visibility of workers' behavior. Because of this remarkable influence relationship, in the daily coal mine safety management, in the prevention of roof accidents, we should pay attention to such unsafe behavior in the process of safety education and training, safety supervision, inspection, etc. [24], as this can effectively reduce the incidence of unsafe behavior.

The significant results of H2 support the positive effect of single-dimensional features on the intervention effect and verify the effectiveness of targeted intervention nodes S1–S7. Bona's view shows that different interventions based on the distribution of risk levels of unsafe behavior among workers can achieve the best results at the lowest managemen<sup>t</sup> cost [45]. Cheng studied the characteristics of occupational accidents in construction enterprises and found that more accidents occurred on the first day of workers' presence in the workplace. According to the results of the time dimension analysis, enterprises can avoid accidents to a grea<sup>t</sup> extent [46]. In the study of human factors in coal mines in China, Chen found that environmental characteristics affected the occurrence of unsafe behavior to some extent. The environmental characteristics mentioned in this study are the spatial distribution of accidents determined by location, working conditions, geological characteristics, etc. It was also shown that mastering the position dimension distribution characteristics of unsafe behavior has a certain role in promoting the safety managemen<sup>t</sup> efficiency of enterprises [47]. In the analysis of the unsafe behavior of the gas explosion accident in China, Yin et al. found that there are grea<sup>t</sup> differences in the frequency of the unsafe behavior of different types of work, which can make safety training easier and more effective [22]. This conclusion is consistent with the results of this paper. Sanmiquel et al. used a database of occupational accident and death reports in Spain's mining industry to analyze the main causes of accidents. Some data mining techniques, such as Bayesian classifiers, decision trees, and contingency tables, were used to discover behavior patterns based on certain rules. The results were helpful in formulating appropriate preventive measures to reduce human injury and death [48]. In the study of road traffic accidents by Kumar, k-mode clustering technology and a correlation analysis algorithm were used to obtain combined characteristics of road traffic accidents. Through the trend analysis of road traffic accidents, it was found that the results of the study have a positive effect on reducing road traffic accidents [25]. The above conclusions are consistent with the results of this paper. The data distribution of dimension information such as risk level, time, position, etc. directly shows the single-dimensional characteristic of insecurity. According to the data distribution of each dimension of unsafe behavior, such dimensions can significantly change the overall safety climate of the enterprise by providing corresponding guidance to the intervention intensity and managemen<sup>t</sup> resource allocation [49]. In the process of practical safety managemen<sup>t</sup> of coal mine enterprises, the single-dimensional feature node should be taken as the basis for providing the guidance and basis for the intervention and resource allocation from the aspects of safety responsibility, safety culture, safety education, safety investment, and so on [50].

The positive results of H3 support the influence prediction between the multi-dimensional features and the intervention effect, which proves that the targeted intervention nodes M1–M6 can effectively improve the effect of behavioral intervention as the multi-dimensional feature nodes in the process of targeted intervention. The multi-dimensional intervention node determined by interaction analysis makes the targeted identification of the intervention objects more specific and the intervention process more precise [51]. The combinatorial localization of multi-dimensional information is a deep mining of unsafe behavior characteristics, which is different from general surface information and reveals the deep-seated characteristics of unsafe behavior [52].

The results of H4–H6 show that the three kinds of targeted intervention nodes also have interactive relationships, and the support of the analysis results for this prediction shows that the practical targeted intervention will be a dynamic process of intervention. When using single-dimensional features and multi-dimensional features to intervene, the key work behavior will change, and the single-dimensional feature information data distribution will also change when the intervention is aimed at multi-dimensional features. Therefore, in the process of targeted intervention in enterprises, intervention nodes are constantly changing. Because targeted intervention nodes are based on data analysis and positioning, this fully reflects the advantages of the data analysis method in the process of security managemen<sup>t</sup> [53]. When implementing targeted intervention in enterprises, it is necessary to collect and update the unsafe behavior data of workers as the intervention continues, and to analyze the data regularly, so as to reposition the unsafe behavior targeted intervention nodes. Finally, in the process of enterprise safety management, the pertinence of worker unsafe behavior intervention is improved, and real-time intervention is realized.
