Development of an HCI-Based Safety Hazards Prediction Model for Indonesia's Electricity Industry

2.1 Occupational Health and Safety (OHS)

Occupational Health and Safety (OHS) is a fundamental requirement for both Micro, Small, and Medium Enterprises (MSMEs) as well as large companies in conducting their business operations. The purpose of implementing Occupational Health and Safety is to maintain the health and safety of the work environment and to protect coworkers, employees' families, consumers, and others who may also be affected by the work environment. This is because OHS is closely related to the outcomes of production. Every industry must be able to reduce the risks of accidents and occupational diseases to prevent decreased production productivity [11].

Occupational Health and Safety (OHS) is crucial for moral, legal, and financial reasons. Every organization has the obligation to ensure that workers and others involved remain in a safe condition at all times. Occupational Health and Safety practices include prevention, sanctions, and compensation, as well as wound healing and care for workers, providing health care, and sick leave. Efforts for safety include conditions that can lead to death, illness, or stress, both in the workplace and at home. Health is related to the employee being free from physical or non-physical illnesses [12].

Work safety is the protection effort carried out by the company to prevent employees from experiencing work-related accidents while performing their tasks. This protection is provided for physical, mental, and social well-being, with preventive measures against health problems or disorders caused by work and environmental factors, as well as common diseases that could cause harm or loss in the workplace. Work safety is the state in which employees are safe and free from accidents while performing their duties. As a result, employees can complete their work as planned and improve their performance [13].

2.2 Safety hazards

Safety hazards (bahaya keselamatan) are potential risks posing threats to safety. Worker safety in the workplace, especially in industries that involve various high-risk activities, is crucial. These hazards can include physical accidents, exposure to hazardous chemicals, or even ergonomic factors that affect workers' long-term health. Safety hazards need to be well-managed to prevent injuries or accidents that could harm both the company and the workers. Therefore, proper identification of potential hazards is essential in creating a safer work environment [14].

Risk management is the primary approach used to identify and manage safety hazards in industries. As an initial step, companies need to conduct a thorough hazard identification, followed by risk analysis and the implementation of mitigation measures. Effective risk management not only reduces the potential hazards but also increases safety awareness among workers. With a deep understanding of the existing risks, preventive measures can be applied to avoid accidents that may occur [14].

Technology also plays an important role in managing safety hazards, especially with advancements in automation systems and data analysis. The use of Human-Machine Interface (HMI) systems, for example, allows operators to monitor operational conditions in real-time and detect potential hazards before they become major issues. Technologies such as automated sensors and data-based monitoring systems can detect risks early and provide quick responses, which in turn helps reduce accidents and improve operational efficiency. These technologies also enable stricter monitoring of high-risk processes [15].

Workplace safety culture also plays a critical role in managing safety hazards. In this regard, creating a strong safety culture encourages workers to be more concerned about their own safety as well as the safety of their colleagues. Organizations that instill safety values in their workplace culture are more likely to succeed in reducing accidents and improving compliance with safety procedures. With heightened awareness and responsibility, workers will be more proactive in attending safety training and following established procedures, thus creating a safer work environment [16].

2.3 Human Computer Interaction

Based on the Encyclopedia Britannica, Human-Computer Interaction (HCI) is an interdisciplinary field that studies how to design optimal interactions between users and computers, as well as developing interfaces that support these interactions. HCI encompasses the communication process between humans and computers in specific ways, using interaction languages to complete particular tasks [17].

As an interdisciplinary field, HCI involves various disciplines such as computer science, psychology, sociology, graphic design, and industrial design. Over time, HCI has transformed from manual activities to web-based interfaces and multimodal intelligent interaction systems. Current research in this field focuses on various aspects, including user personalization, embedded computing, augmented reality, social computing, knowledge-based interaction, emotion-based interaction, and brain-computer interfaces [3]. These developments have had a significant positive impact on the quality of human life.

Human-Computer Interaction (HCI) is a multidisciplinary field that continues to evolve, encompassing various disciplines such as computer science, industrial design, psychology, behavioral science, organizational behavior, and physiology. Since the concept of HCI was first applied to hazard recognition in the construction sector, many new research topics requiring a cross-disciplinary understanding have emerged. However, as the number of scientific publications in this field increases, manual analysis of the literature becomes impractical. The large volume of documents to be reviewed creates a substantial workload, often leading researchers to spend significant time and effort just to identify research focuses and classify the literature correctly. Moreover, the subjective classification process conducted by researchers is highly prone to human error, which can result in discrepancies between the findings in the literature review and the actual conditions [18].

Traditionally, hazard recognition relied on manual monitoring, traditional human resource management, and post-incident analysis. However, modern approaches focus more on risk prediction, accident prevention, deep learning, intuitive devices based on brain waves and eye movements, and multimodal data processing. Therefore, topics arising from the application of human-computer interaction are closely related to technologies such as virtual reality, augmented reality, computer vision, and computer simulation. After formulating the CHR-HCI (Construction Hazard Recognition–Human–Computer Interaction) framework, future research directions can be determined more systematically.

2.4 Safety management in Indonesia

The electricity sector in Indonesia poses significant safety challenges, prompting increased research on effective occupational safety and health (OSH) management strategies. Recent studies emphasize the adoption of formal safety management systems like SMK3 and ISO 45001 as crucial for enhancing workplace safety. Evidence from PT PLN Indonesia Power Barru shows that integrating these frameworks into routine practices helps reduce accidents and improve compliance [19]. The traditional hazard recognization can be seen as below in Figure 1.

1.jpg

Worker engagement and safety awareness are also pivotal. Research at PLTU Bolok Unit II highlights how employee attitudes and knowledge about safety directly affect adherence to safety protocols. This underscores the importance of ongoing training programs to foster a proactive safety culture within the workforce [20].

Risk management approaches have evolved to address both conventional electrical hazards and new challenges such as those emerging during the COVID-19 pandemic. Proactive risk identification and control measures tailored to power plant environments contribute to safer operations [21].

Technological innovation, particularly the use of Internet of Things (IoT) devices, is transforming safety practices by enabling real-time monitoring and quicker hazard response. These advancements align with Indonesia’s national OSH objectives, which aim to reduce workplace incidents through modernized safety solutions.

Analyses of workplace accidents reveal that multiple factors—such as insufficient training, weak safety culture, and inadequate hazard controls—continue to cause incidents. This calls for targeted, data-driven safety interventions specific to the electricity industry’s context [22].

Additionally, studies from related sectors like electronic manufacturing offer valuable safety management methods that can be adapted to electricity operations. PT PLN’s commitment to safety, demonstrated through programs like the Contractor Safety Management System and the goal of “Zero Accidents,” reflects an industry-wide push toward stronger safety governance.

Together, these studies suggest that a comprehensive approach—integrating system implementation, human factors, risk management, technology, and organizational commitment—is essential for improving safety performance and protecting workers in Indonesia’s electricity sector.

2.5 Partial Least Square-Structural Equation Modelling (PLS-SEM)

Partial Least Squares - Structural Equation Modeling (PLS-SEM) is one of the SEM methods frequently used for theory development in exploratory research. PLS-SEM emphasizes explaining variance. In PLS-SEM, constructs are represented by proxies in the form of weighted composites of indicator variables related to the construct. Therefore, PLS-SEM is a composite-based SEM approach, reducing the reliance on the strict assumptions of CB-SEM that all covariation between groups of indicators must be explained by a common factor [23].

PLS-SEM has become increasingly popular in various fields of study due to its ability to handle data that is not normally distributed and highly complex models. PLS-SEM offers several advantages, such as not requiring many assumptions, being suitable for various data scales, and being applicable to small sample sizes. The minimum sample size for PLS-SEM can be determined by two rules: ten times the largest number of indicators used to measure a variable, and ten times the number of independent variables pointing to a dependent variable [23].

PLS-SEM has two main models: the inner model (structural model), which describes the relationships between latent variables, and the outer model (measurement model), which depicts the relationship between manifest variables and latent variables. In the measurement model, PLS uses principal component analysis (PCA) on the variance extraction blocks to observe the relationship between indices and latent variables through total variance, including common variance, specific variance, and total variance. This method falls under confirmatory factor analysis (CFA). The stages of using PLS are as follows [23]:

1. Design the structural model (inner model)
2. Design the measurement model (outer model)
3. Create a path diagram
4. Convert the path diagram into a system of equations
5. Estimate the path diagram, loadings, and weights
6. Evaluate the model
7. Conduct hypothesis testing

The PLS analysis process can be performed using several software tools such as SmartPLS, VPLS, and PLSGUI. In this research, SmartPLS was used for data analysis, as this application allows users to apply measurement scales other than interval scales [23].

At the data collection stage of this study, several results were obtained from interviews, Focus Group Discussions (FGDs), and the collection of research questionnaire data.

4.1 Validity test

At the initial stage of questionnaire distribution, data were collected from 36 respondents, which were subsequently used for the validity and reliability testing of the questionnaire. These tests were conducted to ensure that the strength and clarity of the questionnaire items were well understood by the research respondents. The valiSdity test was performed using correlation analysis with a confidence level of 95%. The results of the questionnaire validity test in this study are presented in Table 2. Based on the validity test results, 42 questionnaire items were found to be valid, while 2 items—namely indicators PC1 and EC5—were found to be invalid. The invalid items were subsequently removed from the research questionnaire.

Table 2. Results of the validity test of the questionnaire

The validity test is conducted to examine whether the indicators of the statements in the measurement tool or questionnaire are valid or not, using the product-moment correlation. If the indicators are valid, they will be used for subsequent research, and if not, those indicators can be disregarded or removed from the questionnaire. The decision of whether the indicators are valid or not is based on the comparison between the calculated r-value (r-hitung) and the critical r-value (r-tabel), or the p-value. An indicator is considered valid if the correlation coefficient r-hitung > r-tabel or the p-value < 0.05. For N=36 (df=N-2=34), the r-tabel value is 0.329. The validity test results show that there are indicators that are invalid, namely PC1 and EC5 in Table 2.

4.2 Reliability test

The reliability test used in this research is the Cronbach's Alpha statistical test. Table 3 shows the results of the reliability test for all indicators in the variables used in this research, which are acceptable. This is in line with Ghozali's statement that a construct or variable is considered reliable if the Cronbach's Alpha value is greater than 0.60. The results indicate that all variables are reliable, as the Cronbach's Alpha values for each variable are greater than 0.60.

Table 3. Reliability test results of the questionnaire

4.3 Structural Equation Modeling-Partial Least Square (SEM-PLS) data processing

The survey data collected in this study, consisting of 203 respondents, were processed and analyzed using the Structural Equation Modeling (SEM) method with a Partial Least Square (PLS) approach. According to Hair et al. [25, 32], PLS is an alternative SEM method used to address complex variable relationships and serves as a highly suitable analytical tool for theory development. In exploratory research such as this, the SEM-PLS method is particularly useful as it helps generate more insightful findings [25]. Furthermore, SEM-PLS is more appropriate in this context because it does not require specific assumptions regarding sample size or data distribution [33]. The software used for the analysis was SmartPLS 4.0.

4.4 Measurement model evaluation

The evaluation of the measurement model involves several requirements that must be met, including convergent validity, reliability, and discriminant validity. A measurement model is considered to meet the criteria for convergent validity when each indicator has an outer loading value above 0.50 and each construct has an Average Variance Extracted (AVE) value of 0.50 or higher [23]. The AVE value represents the average of the squared loadings of a set of indicators associated with a construct. In short, an AVE value of 0.50 indicates that the construct is able to explain more than half of the variance of its indicators [23].

Figure 6 shows that some indicators have outer loading values below 0.50 and some constructs have AVE values less than 0.50, indicating that these constructs do not meet the requirements for convergent validity. In addition, there are constructs with Composite Reliability (CR) and Cronbach’s Alpha values below 0.7, meaning that these constructs do not yet meet the criteria for construct reliability. To fulfill the criteria for convergent validity and reliability, indicators that do not meet the required thresholds need to be removed from the model. The construct reability table can be seen in Tables 4 and 5 as below;

6.jpg

Figure 6. Start SEM-PLS

Table 4. Construct reliability and validity of the initial SEM-PLS model

Table 5. Construct reliability and validity of the initial SEM-PLS model (continued)

After conducting several measurement model evaluations, the final SEM-PLS model was obtained. The iteration process went through three stages to achieve a valid model. Table 6 shows that the outer loading values for each indicator range from 0.584 to 0.904, and the AVE values for each construct range from 0.534 to 0.782, indicating that the model meets the requirements for convergent validity. In addition, Cronbach’s Alpha and Composite Reliability (CR) were used to assess the internal consistency reliability of each construct. Based on Table IV.8, both the Composite Reliability (CR) and Cronbach’s Alpha values exceed 0.7, indicating that the constructs meet the criteria for reliability [23].

Table 6. Reliability and validity of constructs in the final SEM-PLS model

After conducting several measurement model evaluations, the final SEM-PLS model was obtained. The iteration process went through three stages to achieve a valid model. Table 6 shows that the outer loading values for each indicator range from 0.584 to 0.904, and the AVE values for each construct range from 0.534 to 0.782, indicating that the model meets the requirements for convergent validity. In addition, Cronbach’s Alpha and Composite Reliability (CR) were used to assess the internal consistency reliability of each construct. Based on Table IV.8, both the Composite Reliability (CR) and Cronbach’s Alpha values exceed 0.7, indicating that the constructs meet the criteria for reliability [23].

4.5 Discriminant validity testing

Discriminant validity testing was conducted based on the recommendation of Fornell and Larcker [24], which involves comparing the square root of AVE values with the correlations between constructs. Tables 6 and 7 show that the square root of each construct’s AVE is greater than its correlations with other constructs, indicating that the model satisfies the criteria for discriminant validity.

In addition, discriminant validity was also assessed using the HTMT (Heterotrich - Monetarist Ratio) method. The HTMT test calculates the ratio between the correlations across constructs (heterotrich) and the correlations within the same construct (monorail), as estimated from the PLS model. The recommended HTMT threshold is below 0.90 [25]. If the HTMT value between two constructs is less than 0.90, the constructs are considered to have good convergent consistency and can be distinguished from one another. The results in Table IV.10 show that the HTMT values between all construct pairs are below 0.90, leading to the conclusion that all constructs meet the requirements for discriminant validity.

4.6 Hypothesis testing

The structural model evaluation was carried out to test the research hypotheses. The bootstrapping procedure was conducted with 5,000 resamples using SmartPLS 4.0 software [26]. The quality of the structural model was assessed by testing the coefficient of determination (R²), predictive relevance (Q²), and path coefficients [23]. This study tested 11 directional hypotheses using a one-tailed test at a 5% significance level. The last SEM-PLS flowchart can be seen as Figure 7 below:

7.jpg

Figure 7. Last SEM-PLS

The results in Table 8 are summarized as follows:

Hypothesis H1 is accepted, indicating that safety factors have a positive influence on safety climate (β = 0.338, p = 0.000). This means that the better the safety factors, the higher the safety climate.

Hypothesis H2 is rejected, meaning that improvements in personal attributes do not significantly influence safety climate, as indicated by a p-value of 0.137, which is greater than 0.05.

Hypothesis H3 is accepted, showing that accident history has a positive influence on safety climate (β = 0.177, p = 0.004). Better accident history leads to improved safety climate.

Hypothesis H4 is rejected, suggesting that health history does not significantly affect safety climate (p = 0.110 > 0.05).

Hypothesis H5 is also rejected, indicating that personal component does not significantly affect safety climate (p = 0.074 > 0.05).

Hypothesis H6 is accepted, showing that personal component has a positive influence on safety risk (β = 0.179, p = 0.014). This means that stronger personal components are associated with higher safety risk.

Hypothesis H7 is accepted, indicating that environmental component positively affects safety risk (β = 0.277, p = 0.000). Better environmental components are linked to increased safety risk.

Hypothesis H8 is rejected, suggesting that organizational component does not significantly affect safety risk (p = 0.262 > 0.05).

Hypothesis H9 is accepted, indicating that safety climate positively influences safety hazards (β = 0.581, p = 0.000). A higher safety climate leads to an increase in safety hazards.

Hypothesis H10 is rejected, showing that safety climate does not significantly affect safety risk (p = 0.112 > 0.05).

Hypothesis H11 is also rejected, meaning that safety hazards do not significantly affect safety risk (p = 0.126 > 0.05).

Hypothesis H1 is accepted, indicating that safety factors have a positive influence on safety climate (β = 0.338, p = 0.000). This means that the better the safety factors—such as availability of safety equipment, training programs, clear safety procedures, effective supervision, and open communication—the higher the safety climate in the workplace. Safety factors measured by indicators SF1, SF2, SF3, SF4, and SF5 significantly contribute to creating a safer work environment by promoting employees’ perception of organizational commitment to safety and encouraging safe behavior.

From a theoretical perspective, this finding aligns with the Safety Culture Theory and Organizational Climate Theory, which emphasize that tangible safety measures and management’s visible commitment to safety are critical in shaping employees’ shared perceptions about workplace safety. When employees see that safety resources and protocols are adequately provided and consistently enforced, they develop trust and confidence in the organization’s safety priorities. This positive safety climate fosters collective responsibility and motivates individuals to adhere to safety standards, reducing workplace accidents and injuries.

Practically, this result highlights the importance of continuous improvement in safety infrastructure, training quality, and communication channels to sustain a positive safety climate. Organizations should invest in regular safety audits, refreshers, and employee engagement to ensure these safety factors are effectively maintained and perceived by all workers.

Hypothesis H2 is rejected, meaning that improvements in personal attributes do not significantly influence safety climate, as indicated by a p-value of 0.137, which is greater than the significance threshold of 0.05. Personal attributes measured by indicators such as age, work experience, and last education level do not have a significant effect on safety climate.

This suggests that safety climate is predominantly shaped by organizational factors and safety management practices rather than by individual employee characteristics. According to Social Cognitive Theory, safety climate is a collective perception influenced mainly by shared experiences, leadership behavior, and organizational policies rather than demographic differences among employees. Individual factors like age or education may affect personal behavior or risk perception but do not substantially alter the overall workplace safety climate.

The final SEM-PLS Flowchart can be seen in Figure 8 as below:

Table 7. Reliability and validity of constructs in the final SEM-PLS model (continued)

Table 8. Hypothesis testing results

8.jpg

Figure 8. Final T-statistic values of SEM-PLS

From a practical standpoint, this finding underscores the need for organizations to focus their safety efforts on systemic interventions, such as improving management commitment, strengthening safety communication, and implementing consistent safety policies, instead of relying on employee demographics to predict or improve safety climate. By targeting organizational-level changes, companies can create an inclusive and effective safety environment that benefits all employees regardless of their individual backgrounds.

4.7 Predictive model

The level of variance in the endogenous variables that can be explained by the exogenous variables is overall explained by R². According to Cohen (1988), there are three criteria for R² values: weak (R² ranges from 0.02 to 0.13), moderate (R² ranges from 0.13 to 0.26), and substantial (R² greater than 0.26). Based on Table 9 the model explains 28.9% of the variance in safety climate, 33.8% of the variance in safety hazards, and 17.3% of the variance in safety risk, with the remainder explained by variables outside the model. The results of this study indicate that the safety climate and safety hazards models fall into the substantial category, while the safety risk model falls into the moderate category.

Next, using the blindfolding method to assess the predictive relevance (Q²), which measures how well the observed values are predicted by the model and its parameter estimates, Table 9 shows Q² values ranging from 0.076 to 0.308. A Q² value greater than 0 indicates that the model has good predictive relevance [26].

Table 9. Amount R² and Q²

The results of this study provide a comprehensive view of the critical factors that influence workplace safety in Indonesia’s electricity industry, particularly in distribution units. The structural equation modeling (SEM-PLS) analysis confirms the significant influence of personal competence (PC), safety climate (SC), equipment condition (EC), and management commitment (MC) on safety hazard risk.

These findings are consistent with the study by Baby et al. [9] which highlighted how personal and organizational factors—such as competence and safety awareness—directly correlate with the likelihood of workplace accidents. The current research reinforces this by demonstrating that personal competence and safety climate are among the strongest predictors of safety hazard levels. The model further reveals that management commitment mediates the influence of other variables, underscoring the importance of leadership in fostering a culture of safety.

The significance of equipment condition (EC) also aligns with findings by Castillo-Rosa et al. [7] who emphasized the role of technical compliance and regular maintenance in preventing high-severity accidents. However, in the Indonesian context, limited resources and uneven regulatory enforcement make consistent equipment maintenance a continuing challenge. Therefore, the study emphasizes the need for clear technical standards and centralized asset audits.

From a practical perspective, this model has been translated into a KPI-based safety manual that can be integrated into PLN’s distribution units. This manual is designed to assist field supervisors in monitoring, evaluating, and improving worker behavior and hazard mitigation in real-time. Unlike traditional static manuals, this KPI approach allows dynamic adjustment based on measurable performance indicators.

Enhance Technical and Behavioral Training: Based on the significance of personal competence, electricity companies should prioritize targeted training that includes both technical certification and behavioral safety simulations. and strengthen management engagement with leadership training should be expanded for managerial roles to foster a more proactive safety climate. This includes regular feedback sessions, site visits, and worker recognition programs.