Human Reliability Analysis in Interaction Design Based on CREAM, FCE, and DEMATEL

Abstract

Human reliability analysis of the human–computer interaction process between users and systems is critical because human error can introduce significant system risks. Interaction systems designed with human reliability analysis can reduce human error. This study proposed a research methodology for analyzing human error to design interactive systems that align with users’ cognitive demands. First, the cognitive reliability and error analysis method (CREAM) is used to investigate cognitive function failures and determine the nominal cognitive failure probability. Next, fuzzy comprehensive evaluation (FCE) is used to assess the level of common performance conditions (CPCs). Subsequently, the decision-making trial and evaluation laboratory (DEMATEL) method is employed to compute the factor centrality weights of CPCs and human intrinsic factors (HIFs). The interactions among CPCs are analyzed, leading to the determination of cognitive impact weights. Then, the cognitive failure probability is calculated by combining factor centrality weights and cognitive impact weights. Finally, error causes are analyzed to propose optimization strategies and implement design improvements. An in-vehicle information system was used to validate the proposed approach. The findings revealed that this method effectively minimizes cognitive failure probability during system interaction. It also identifies the causes of human error in human–computer interactions and offers a systematic strategy to enhance human reliability in interaction design.

1. Introduction

With the advancement of information technology, the human–computer interaction process has become more complex [1], leading to a higher susceptibility to human error [2]. Human error in interactive systems is an inherent aspect of system performance [3]. Interaction design for human error can effectively improve the usability of interactive systems [4,5]. Human reliability analysis of interactive systems can reduce human error by providing recommendations for design improvements, thereby minimizing the negative impact of these errors and improving system usability [6,7,8].

Human reliability analysis (HRA) provides insights into human–computer interaction behaviors during tasks and quantifies failures [9]. The first generation of HRA methods is recognized for its intuitiveness and practicality [10]. However, the increasing demand for a deeper understanding of human–computer interaction has driven the development of second-generation HRA methods [11]. The cognitive reliability and error analysis method (CREAM) is a representative second-generation HRA method [12]. It is known for identifying cognitive tasks and determining the causes of cognitive reliability degradation [13]. Existing studies have applied the CREAM to analyze human reliability in system operations [14]. For example, Chen et al. [15] used the CREAM to assess human error in the operation of high-speed railway systems. Additionally, the CREAM has been employed in accident risk analysis. For instance, Elidolu et al. [16] used the CREAM to quantify the causes of failures in a cruise ship explosion at sea. However, to the best of the authors’ knowledge, limited research has been conducted on applying the CREAM in interaction design to analyze cognitive function failures.

The CREAM defines the factors influencing cognitive, technological, environmental, and organizational performance as common performance conditions (CPCs) [17]. However, two key limitations encountered in the practical application of the traditional CREAM include the objective assessment of the level of CPCs and the detailed analysis of the interaction among CPCs [18]. The traditional CREAM lacks refined evaluation factors for CPCs [19], and the evaluation process relies heavily on expert knowledge and experience, which are often subjective and inconsistent [20]. Fuzzy comprehensive evaluation (FCE) is a widely used uncertainty assessment method that improves the objectivity of assessment results by representing expert judgments using fuzzy linguistic terms [21]. By simplifying complex decision-making processes, FCE provides a reliable and adaptable tool for addressing ambiguity in evaluation scenarios [22]. It has been demonstrated to be useful in optimizing evaluation effectiveness [23,24]. Therefore, to mitigate the impact of subjective expert assessments on the quantification of human reliability, this paper integrates FCE into the traditional CREAM framework to evaluate the levels of CPCs in interactive systems.

The traditional CREAM primarily examines the effects of CPCs on cognitive reliability in the cognitive dimension, providing only a simplified analysis of the interactions between factors [25]. However, the interrelationships among CPCs can affect their impact on interactive systems, influencing the results of human reliability analysis [26]. Therefore, a correlation impact analysis of CPCs is crucial for interaction systems. The decision-making trial and evaluation laboratory (DEMATEL) method effectively evaluates complex systems by analyzing factor relationships [27]. It has been extensively applied to investigate the interrelationships among factors that influence system security [28]. For example, Shi et al. [29] used the DEMATEL method to analyze the interdependence and interactions of the factors influencing the cargo loading process. DEMATEL improves the reliability of failure analysis in interactive systems by identifying the causal relationships and interactions among factors [30]. Therefore, to quantify the impact of human–computer interaction factors on the cognitive reliability assessment of interactive systems, DEMATEL is incorporated in this paper.

Human intrinsic factors (HIFs) influence the user’s comprehension and execution of tasks, which is crucial for determining the human reliability of interactions with the system [31]. Considering the impact of HIFs on user behavior in interaction design can enhance interactive quality [32]. The traditional CREAM, which does not explicitly incorporate the influence of HIFs on cognitive function failures, may have limited accuracy in predicting human error in interactive systems [19]. Therefore, to provide a more comprehensive assessment of cognitive function failures, this paper integrates the impact of CPCs and HIFs.

Building on the above analysis of DEMATEL’s role and the integration of CPCs and HIFs, this study offers distinct research novelties. The novelty of this study lies in the synergistic effect of four dimensions: first, a human–computer interaction-specific CPC/HIF analysis system, which has a wide range of applications and is highly suitable for human reliability analysis; second, a normalized product-based weight integration mechanism, which solves the problem of low fusion accuracy; third, a closed-loop design improvement cycle, which realizes iterative optimization of human–computer interaction design; and finally, an end-to-end workflow, which covers the entire link from indicator establishment to effect verification.

This paper introduces a method for human reliability analysis in interaction design, integrating the CREAM, FCE, and DEMATEL. This research includes three main contributions that enhance the applicability and accuracy of human reliability analysis in interactive systems. First, to reduce uncertainty in evaluating CPC levels, FCE is incorporated. Second, to comprehensively explore the impact of human error in interactive systems, the impact of CPCs and HIFs is integrated into the traditional CREAM framework. Third, to enhance the accuracy of cognitive reliability analysis, the interrelationships among CPCs are analyzed, and the factor centrality weights of CPCs and HIFs are calculated using the DEMATEL method.

The structure of the paper is as follows: Section 2 outlines the research background, while Section 3 explains the proposed methodology. Section 4 presents a case study, followed by a discussion of its results in Section 5. Finally, Section 6 provides the conclusions.

2. Theoretical Background

2.1. Cognitive Reliability and Error Analysis Method

The CREAM was proposed by Hollnagel in 1998 [33]. It integrates human cognitive, technological, environmental, and organizational factors into a reliable cognitive model and error classification system [34]. The model emphasizes the influence of cognitive processes and environmental factors on human error, referring to these as CPCs [35]. The CREAM has been widely applied in human reliability analysis in fields such as nuclear power [36], aerospace [37], and the maritime sector [38].

The traditional CREAM consists of basic and extended methods. The basic CREAM assesses the overall perceived probability of failure for the entire system, and the extended CREAM calculates the cognitive failure probability (CFP) for each step [39]. The CREAM is applicable for both predictive analysis of complex human–machine systems to identify tasks prone to human error, and retrospective analysis to identify causative factors that lead to reliability degradation [40]. In this study, the extended CREAM is applied to analyze the CFPs of users in completing human–computer interaction tasks and to explore the root causes of human error.

2.2. Fuzzy Comprehensive Evaluation

FCE is a widely applied decision-making approach that combines fuzzy logic and multiple-factor evaluation techniques to address complex and uncertain systems [41]. This method converts qualitative information into measurable data and integrates various evaluation criteria into a unified assessment framework by constructing a fuzzy evaluation matrix [42]. Due to its flexibility and robustness, the FCE method has been extensively used in program assessment [43], risk analysis [44], and performance evaluation [24], providing a systematic and reliable framework for decision-making in uncertain conditions.

The traditional CREAM determines the level of CPCs through expert evaluation, but expert assessments often involve uncertainty. To reduce the uncertainty in expert evaluations and enhance the CREAM’s effectiveness in interaction design, this research uses FCE to refine the evaluation process by synthesizing experts’ ratings of the evaluation factors through fuzzy logic.

2.3. Decision-Making Trial and Evaluation Laboratory

DEMATEL was developed by the Battelle Memorial Institute in Geneva in the 1970s [45]. It has been used to address complex system problems involving the interaction of multiple factors, providing decision-makers with a clear understanding of the system dynamics [28]. The DEMATEL method has been successfully applied to analyze factor correlations and importance across various domains [46,47]. In this paper, DEMATEL is used to improve the accuracy of human reliability analysis in interactive systems by quantifying the factor centrality weights of CPCs and HIFs on cognitive function failures. In addition, to reduce the limitation of the traditional CREAM whereby it does not thoroughly analyze the interaction of CPCs, this paper uses DEMATEL to analyze the interrelationships among CPCs in interactive systems.

3. Proposed Methodology

This paper proposes a novel method integrating the CREAM, FCE, and DEMATEL to analyze human error in interaction design and reduce cognitive load on users interacting with the system. The framework of the proposed methodology is shown in Figure 1 and consists of five stages. In stage 1, cognitive function failures are evaluated. In stage 2, CPC levels are assessed. In stage 3, the HIF and CPC weights are calculated. In stage 4, the cognitive failure probability is determined. In stage 5, recommendations for design improvements are proposed.

3.1. Analysis of Cognitive Function Failures

3.1.1. Conducting Hierarchical Task Analysis and Establishing a Cognitive Demand Profile

To analyze the tasks of the target interactive system, hierarchical task analysis (HTA) was performed. HTA provides a clear, structured breakdown of complex tasks [48]. It is a widely used method for decomposing tasks into subtasks and represents the result of a hierarchical tree diagram [49].

A cognitive demand profile is constructed based on the user’s performance during the completion of subtasks from the hierarchical tree diagram. The key cognitive activities and associated cognitive functions involved in interacting with the system are identified through the observation and recording of task completion. The fifteen cognitive activity types and four cognitive functions used to construct the cognitive demand profile are presented in

Table 1. The elements of the CREAM cognitive demand profile.

Cognitive Activity Type	Cognitive Functions
	Observation	Interpretation	Planning	Execution
Co-ordinate			+	+
Communicate				+
Compare	+	+
Diagnose		+	+
Evaluate		+	+
Execute				+
Identify	+	+
Maintain			+	+
Monitor	+	+
Observe	+
Plan			+
Record		+		+
Regulate	+			+
Scan	+
Verify	+	+

Note: The “+” symbol denotes a correlation between cognitive activities and functions [18].

.

3.1.2. Obtaining the Nominal Cognitive Failure Probability

The CREAM categorizes four types of cognitive functions into various cognitive function failures and assigns the corresponding nominal CFPs, as shown in

Table 2. Cognitive function failures and nominal CFPs in CREAM.

Cognitive Functions	Cognitive Function Failures	Nominal CFPs
Observation	O1 (Wrong object observed)	0.001
	O2 (Wrong identification)	0.07
	O3 (Observation not made)	0.07
Interpretation	I1 (Faulty diagnosis)	0.2
	I2 (Decision error)	0.01
	I3 (Delayed interpretation)	0.01
Plan	P1 (Priority error)	0.01
	P2 (Inadequate plan)	0.01
Execution	E1 (Action of wrong type)	0.003
	E2 (Action of wrong time)	0.003
	E3 (Action on wrong object)	0.0005
	E4 (Action out of sequence)	0.003
	E5 (Missed action)	0.03

[19]. Human error is analyzed based on the cognitive demand profile, and the cognitive function failures occurring during system interaction are identified.

3.2. Assessment of CPC Levels

Interaction contexts are established by analyzing CPCs defined based on the target interaction system [37]. By determining CPC levels, interaction contexts where human error occurs during human–computer interaction can be identified. This paper evaluates the impact level of each CPC using FCE. Each CPC is further divided into sub-factors. The system’s performance level is then derived through FCE, which involves the following five steps:

Step 1: Establish the factor set.

The factor set is a collection of factors that affect the level of each CPC, represented by set C, $C=\left\{c_1,c_2,\cdots ,c_i,\cdots ,c_m\right\}$, and $c_i(i=1,2,\cdots ,m)$ represents the refined sub-factors for each CPC.

Step 2: Establish the weight set.

The weight set reflects the importance of each sub-factor in the CPCs by assigning weight $a_i(i=1,2,\cdots ,m)$ to sub-factor c_i. The weights of all sub-factors form set A, $A=\left\{a_1,a_2,\cdots ,a_i,\cdots ,a_m\right\}$, where ${\displaystyle \sum _{i=1}^m{a}_i=1}$, with $a_i\ge 0$.

Step 3: Establish the evaluation set.

The evaluation set is a collection of evaluation results of the CPCs of the interactive system. The evaluation set is set as L, $L=\left\{l_1,l_2,\cdots ,l_j,\cdots ,l_n\right\}$, and $l_j(j=1,2,\cdots ,n)$ denotes the impact level possessed by the sub-factors of the CPCs.

Step 4: Perform a fuzzy comprehensive evaluation.

The evaluation is conducted on the ith sub-factor c_i of the CPCs, with the affiliation degree of the CPCs to the jth level l_j in the evaluation set denoted as r_ij. The evaluation results of the sub-factors for each CPC are then aggregated into an evaluation matrix R, as shown in Equation (1).

$$R=\left[\begin{array}{cccccc}{r}_{11}& r_{12}& \cdots & r_{1j}& \cdots & r_{1n}\\ r_{21}& r_{22}& \cdots & r_{2j}& \cdots & r_{2n}\\ ⋮& ⋮& & ⋮& & ⋮\\ r_{i1}& r_{i2}& \cdots & r_{ij}& \cdots & r_{in}\\ ⋮& ⋮& & ⋮& & ⋮\\ r_{m1}& r_{m2}& \cdots & r_{mj}& \cdots & r_{mn}\end{array}\right]$$

(1)

Based on the weight set A and evaluation matrix R, the fuzzy synthesized evaluation set B is calculated using Equations (2) and (3).

$$B=A\circ R=\left(b_1,b_2,\cdots ,b_j,\cdots ,b_n\right)$$

(2)

$$b_j={\displaystyle \sum _{i=1}^m{a}_i{r}_{ij}}$$

(3)

where $b_j(j=1,2,\cdots ,n)$ is the fuzzy synthesized evaluation index and “$\circ$” represents the fuzzy synthesis process.

Step 5: The evaluation indicators are processed using the maximum affiliation method to determine CPC levels.

3.3. Calculation of HIF and CPC Weights

In this paper, to calculate the factor centrality weights of CPCs and HIFs, DEMATEL is employed. This method quantifies the relationships between factors and their respective degrees of influence with precision. DEMATEL quantitatively analyzes the relationships between factors to identify their causality and centrality, typically following five steps:

Step 1: Define and evaluate factors.

Identify the factors influencing the system and assess the relationships between factors using expert judgment. These relationships are categorized into four levels of influence: 0 represents no influence, 1 represents slight influence, 2 represents moderate influence, and 3 represents strong influence.

Step 2: Obtain the direct relationship matrix X.

Based on the evaluations from Step 1, the factors are compared pairwise to assess their influence. The arithmetic mean of the scores is used to construct an n × n direct relationship matrix, denoted as $X={\left[x_{ij}\right]}_{n\times n}$, where x_ij represents the degree to which factor i influences factor j, with all diagonal elements set to 0.

Step 3: Calculate the normalized direct relation matrix D.

The normalized direct relation matrix D can be calculated using Equations (4) and (5).

$$Q=\min \left[{\displaystyle \frac{1}{\underset{1\le i\le n}{\max }\left({\displaystyle \sum _{j=1}^n{x}_{ij}}\right)}}\right.,\left.{\displaystyle \frac{1}{\underset{1\le j\le n}{\max }\left({\displaystyle \sum _{i=1}^n{x}_{ij}}\right)}}\right]$$

(4)

$$D=X\times Q$$

(5)

Step 4: Calculate the total impact relationship matrix T.

Due to $\underset{k\to \infty }{\lim }{D}^k=0$, the total matrix $T={\left[t_{ij}\right]}_{n\times n}$ can be calculated using Equation (6).

$$T={\displaystyle \sum _{k=1}^{\infty }{D}^k}={\displaystyle \frac{D}{I-D}}$$

(6)

where I is the identity matrix, 0 is the null matrix, and t_ij represents the degree to which factor i affects factor j.

Step 5: Conduct an analysis to get the sum of rows and columns.

Let h and g denote the sum of the rows and columns in matrix T, which can be calculated using Equations (7) and (8).

$$h=h_{n\times 1}={\left[{\displaystyle \sum _{j=1}^n{t}_{ij}}\right]}_{n\times 1}$$

(7)

$$g=g_{n\times 1}={\left[{\displaystyle \sum _{i=1}^n{t}_{ij}}\right]}_{1\times n}^t$$

(8)

Let h_i represent the sum of the ith row and g_i represent the sum of the ith column. In this paper, h_i indicates the degree to which CPC i influences other CPCs, while g_i represents the degree to which it is influenced by others. The value of h_i + g_i represents the degree of centrality of CPC i, reflecting its relative importance. Therefore, the value of h_i + g_i can then be normalized to calculate the factor centrality weight of CPC i using Equation (9).

$$w_i={\displaystyle \frac{h_i+g_i}{{\displaystyle \sum _{i=1}^9\left(h_i+g_i\right)}}}$$

(9)

where w_i is the factor centrality weight of CPC i on subtasks.

In this paper, h_j indicates the degree to which HIF j influences other HIFs, while g_j represents the degree to which it is influenced by others. Based on the analysis above, the value of h_j + g_j can then be normalized to calculate the factor centrality weight of HIF j using Equation (10).

$${\nu }_j={\displaystyle \frac{h_j+g_j}{{\displaystyle \sum _{j=1}^9\left(h_j+g_j\right)}}}$$

(10)

where v_j is the impact weight of HIF j on subtasks.

Based on the DEMATEL analysis results, we identify the CPCs influenced by others as CPCs with values higher than the average. Then, the relevance adjustment rules are applied to analyze the interactions among CPCs. The cognitive impact weights are determined based on CPC interactions and levels.

3.4. Calculation of the Cognitive Failure Probability

The factor centrality weight of CPCs and the impact weight of HIFs are integrated via Equation (11) to yield the comprehensive weight.

$$p_i={\displaystyle \frac{u_i{w}_i}{v_j}}$$

(11)

where p_i is the comprehensive weight, u_i is the cognitive impact weight of the ith CPC, w_i is the factor centrality weight of CPC i on subtasks, and v_j is the impact weight of HIF j on subtasks.

Based on the derived comprehensive weight, the CFP for each subtask is calculated via Equation (12).

$$CFP=CF{P}_0{\displaystyle \prod _{i=1}^m{p}_i}$$

(12)

where CFP₀ denotes the nominal CFP of each subtask.

Note that Equation (11) combines the node centrality and the factor impact degree of cognitive risk factors within the DEMATEL causal network. The normalized product form is adopted to eliminate dimensional differences and highlight their synergistic effect, which is more reasonable than additive or unnormalized product forms, and this combination is a fusion operator rather than a weight update or risk amplification operator. Equation (12), which is closely related to Equation (11) in the cognitive reliability analysis process, uses multiplicative aggregation for CFP. This method assumes the conditional independence of cognitive failure events across different factors within the human–computer interaction scenario. To reconcile DEMATEL-based interaction modeling with multiplicative aggregation that may implicitly assume separability, we adopt a two-stage strategy: first using DEMATEL to screen key cognitive factors, and then applying multiplicative aggregation to calculate CFP. The key cognitive factors screened by DEMATEL inherently have independent characteristics in terms of their impact on cognitive failure, which ensures the rationality of the multiplicative aggregation method and effectively resolves potential conflicts between the two methods.

3.5. Recommendations for Design Improvements

To improve the design, we recommend the following: Identify human-error-prone subtasks in the interactive system based on the CFPs. Analyze the causes of human error using CREAM’s cause-effect matrix proposed by Hollnagel [33]. Develop optimization strategies and implement design improvements for the target interactive system [50].

4. Case Study

The increasing complexity of human–machine interactions in modern in-vehicle information systems (IVIS) underscores the importance of researching human reliability [51]. An IVIS requires multitasking and intensive cognitive processing, which can overwhelm drivers, particularly under complex driving conditions [52]. This paper focuses on drivers’ interaction errors with IVISs and identifies cognitive function failures. Strategies for improving the design of interactive systems are proposed based on the analysis. The interface of the target IVIS is illustrated in Figure 2.

The focus group for this case study consisted of ten members, including four reliability experts and six interaction designers. All reliability experts possessed more than ten years of professional experience, while each interaction designer had over seven years of relevant professional experience. This study recruited 30 drivers (14 females and 16 males, aged 30–60 years) to interact with the IVIS. All participants had at least five years of experience using such systems.

4.1. Analysis of Cognitive Function Failures in Drivers Interacting with IVIS

The team conducted the HTA of the in-vehicle navigation function, resulting in a hierarchical tree diagram representing the tasks performed by drivers during navigation. The results are presented in Figure 3.

A cognitive demand profile for drivers performing in-vehicle navigation tasks was established, and the result is shown in

Table 3. Cognitive demand profile for drivers using the in-vehicle navigation function.

Subtasks	Cognitive Activity	Cognitive Functions
		Observation	Interpretation	Planning	Execution
1.1	Observe	+
1.2.1	Identify	+	+
1.2.2	Co-ordinate			+	+
1.2.3	Regulate	+			+
1.3.1	Execute				+
1.3.2	Observe	+
1.3.3.1	Identify	+	+
1.3.3.2	Co-ordinate			+	+
1.3.4	Observe	+
1.4.1	Execute				+
1.4.2	Observe	+

. For example, the key cognitive activity for task 1.2.1 (Enter the destination name) is to identify, and the associated cognitive functions are observation and interpretation.

Based on the cognitive demand profile analysis, human error and cognitive function failures during drivers’ interactions with the IVIS were identified. The nominal CFPs were derived, and the results are shown in

Table 4. Cognitive function failures that occurred in the interaction with the IVIS.

Subtasks	Cognitive Function Failures	Nominal CFPs
1.1 Click on the navigation icon	O2 (Wrong identification)	0.07
1.2.1 Enter the destination name	I3 (Delayed interpretation)	0.01
1.2.2 Select a frequently used destination	P1 (Priority error)	0.01
1.2.3 Choose a navigation route	E1 (Action of wrong type)	0.003
1.3.1 Click the “Start” button	E3 (Action on wrong object)	0.0005
1.3.2 View navigation information	O3 (Observation not made)	0.07
1.3.3.1 Change the destination	I2 (Decision error)	0.01
1.3.3.2 Add waypoints	P2 (Inadequate plan)	0.01
1.3.4 Return and continue navigation	O2 (Wrong identification)	0.07
1.4.1 Click the “Exit” button	E1 (Action of wrong type)	0.003
1.4.2 Click the home icon	O2 (Wrong identification)	0.07

.

4.2. Assessment of CPC Levels Related to IVIS

The CPCs of the traditional CREAM are not fully applicable to human reliability analysis in automobile driving, requiring adjustments to reflect human–computer interaction. Nine CPC factors and sub-factors relevant to drivers’ use of the IVIS were identified by analyzing the automotive driving context [53,54,55,56] and human–computer interaction [57,58,59,60], and the results are shown in

Table 5. Factors and sub-factors of CPCs for drivers using the IVIS.

CPC Factors	CPC Sub-Factors
CPC1 (Degree of visualization of the interface)	CPC1-1 (Information hierarchy)
	CPC1-2 (Visual fluency)
	CPC1-3 (Icon recognition)
	CPC1-4 (Color contrast)
CPC2 (Rationalization of interaction mechanisms)	CPC2-1 (Logic clarity)
	CPC2-2 (Ease of operation)
	CPC2-3 (Function matching)
	CPC2-4 (Speed of response)
CPC3 (Adequacy of information feedback)	CPC3-1 (Timeliness of feedback)
	CPC3-2 (Clarity of information)
	CPC3-3 (Consistent format)
	CPC3-4 (Comprehensiveness of feedback)
CPC4 (Number of simultaneous tasks)	CPC4-1 (Task prioritization)
	CPC4-2 (Parallel operation)
	CPC4-3 (Switching methods)
	CPC4-4 (Task focus)
CPC5 (The cognitive load of drivers)	CPC5-1 (Transient cognition)
	CPC5-2 (Load distribution)
	CPC5-3 (Dynamic information)
	CPC5-4 (Operational stability)
CPC6 (Personal health status)	CPC6-1 (Vision health)
	CPC6-2 (Hearing health)
	CPC6-3 (Reactivity)
	CPC6-4 (Energy status)
CPC7 (Available task execution time)	CPC7-1 (Attention allocation)
	CPC7-2 (Energy management)
	CPC7-3 (Time constraints)
	CPC7-4 (Time perception)
CPC8 (Cockpit environment of automobiles)	CPC8-1 (Space layout)
	CPC8-2 (Noise control)
	CPC8-3 (Temperature control)
	CPC8-4 (Wide field of vision)
CPC9 (Driving skills and experience)	CPC9-1 (Driving skills)
	CPC9-2 (Road awareness)
	CPC9-3 (Driving experience)
	CPC9-4 (Adaptability to road conditions)

. These CPCs were categorized into multiple levels based on the traditional CREAM and presented in

Table 6. The level indicators of the nine CPCs.

Factors	Level
CPC1	Supportive; Adequate; Tolerable; Inappropriate
CPC2	Very sensible; Reasonable; Tolerable; Unreasonable
CPC3	Very efficient; Efficient; Inefficient; Deficient
CPC4	Appropriate; Acceptable; Inappropriate
CPC5	Low; Medium; High
CPC6	Healthy; Generic; Unhealthy
CPC7	Adequate; Limited; Inadequate
CPC8	Appropriate; Acceptable; Inappropriate
CPC9	Enough; Limited; Insufficient

.

We then evaluated the level of CPCs of the target IVIS using FCE. Take CPC1 (Degree of visualization of the interface) as an example. According to Table 5, the factor set C₁ = {Information hierarchy, Visual fluency, Icon recognition, Color contrast} was established. The team members were invited to rate the CPC sub-factors on a scale from 1 to 10, where higher scores indicate greater importance. The expert assessments were collected using open group discussions. The group for this case study consisted of ten members, including four reliability experts and six interaction designers. The members conducted joint open discussions to evaluate the sub-factors on a 1–10 scale, and multiple rounds of discussion were conducted to ensure the consistency and reliability of the assessment results. The total scores for each sub-factor were normalized to calculate their weights using Equation (13), and the results are presented in

Table 7. Scoring and assessment weights for CPC sub-factors.

Sub-Factors	M1	M2	M3	M4	M5	M6	M7	M8	M9	M10	Total	Weights
CPC1-1	6	7	5	4	6	6	5	8	8	9	64	0.247
CPC1-2	5	6	6	7	7	7	4	7	6	7	62	0.239
CPC1-3	9	7	8	8	6	6	7	8	9	7	75	0.290
CPC1-4	5	8	7	7	6	5	6	7	4	3	58	0.224
CPC2-1	6	8	9	7	5	5	4	5	7	6	62	0.238
CPC2-2	9	8	8	8	9	6	6	7	6	8	75	0.289
CPC2-3	6	7	6	7	5	7	5	5	8	8	64	0.246
CPC2-4	5	5	5	4	4	6	7	7	7	9	59	0.227
CPC3-1	7	6	6	9	9	9	7	8	8	6	75	0.284
CPC3-2	7	6	6	8	8	8	8	9	9	5	74	0.281
CPC3-3	4	5	5	5	5	6	6	7	6	7	56	0.212
CPC3-4	4	5	6	8	8	6	5	5	5	7	59	0.223
CPC4-1	5	7	7	6	7	6	5	4	4	8	59	0.242
CPC4-2	5	6	6	4	6	5	7	6	8	4	57	0.234
CPC4-3	8	4	3	6	3	7	4	8	8	6	57	0.233
CPC4-4	6	6	8	8	9	8	8	6	6	6	71	0.291
CPC5-1	4	7	5	5	9	8	5	9	5	8	65	0.248
CPC5-2	5	6	7	7	8	6	8	7	5	7	66	0.252
CPC5-3	8	9	6	8	8	7	6	7	6	6	71	0.271
CPC5-4	7	6	6	6	3	8	7	4	6	7	60	0.229
CPC6-1	8	7	8	5	7	7	5	7	8	6	68	0.246
CPC6-2	8	5	7	7	8	7	5	6	7	7	67	0.242
CPC6-3	8	8	9	6	8	6	7	7	8	8	75	0.272
CPC6-4	9	7	6	5	5	8	5	9	6	6	66	0.240
CPC7-1	7	5	6	6	7	8	6	9	7	9	70	0.276
CPC7-2	5	7	8	6	8	7	6	6	9	8	70	0.275
CPC7-3	7	6	4	4	5	6	3	6	5	9	55	0.217
CPC7-4	6	5	6	5	5	8	4	6	7	7	59	0.232
CPC8-1	5	7	7	6	6	5	5	5	5	8	59	0.263
CPC8-2	4	7	5	6	4	6	6	4	5	8	55	0.246
CPC8-3	4	4	6	5	6	4	5	3	3	7	47	0.210
CPC8-4	6	7	7	7	5	6	5	6	8	6	63	0.281
CPC9-1	7	7	8	7	6	8	8	6	6	8	71	0.278
CPC9-2	5	6	6	6	8	7	7	3	7	6	61	0.239
CPC9-3	6	5	7	8	4	5	4	4	5	5	53	0.208
CPC9-4	8	8	8	7	6	7	6	6	7	7	70	0.275

.

$$f_i={\displaystyle \frac{{\displaystyle {\sum }_{k=1}^m{s}_{ik}}}{{\displaystyle {\sum }_{j=1}^n{\displaystyle {\sum }_{k=1}^m{s}_{jk}}}}}$$

(13)

where f_i represents the normalized weight of the sub-factor i, and s_ik denotes the score given by the kth member to sub-factor i. The total number of members is m, and n represents the total number of sub-factors.

The set of weights for the evaluation factors was established according to Table 7 as A₁ = (0.247, 0.239, 0.290, 0.224). The evaluation set for CPC1 was then built according to Table 7 as L₁ = {Supportive, Adequate, Tolerable, Inappropriate}. The evaluation matrix R₁ of CPC1 was constructed according to Equation (1). For example, analyzing the affiliation degrees of R₁ shows that for the Information hierarchy level of CPC1, 10% of members considered it supportive, 10% considered it adequate, 20% considered it tolerable, and 60% considered it inappropriate.

$$R_1=\left[\begin{array}{cccc}0.1& 0.1& 0.2& 0.6\\ 0& 0.3& 0.3& 0.4\\ 0.1& 0.1& 0.4& 0.4\\ 0& 0.2& 0.3& 0.5\end{array}\right]$$

Using the weight set A₁ and evaluation matrix R₁, fuzzy synthesis was applied to the evaluation results of CPC1. The fuzzy comprehensive evaluation set B₁ was obtained using Equations (2) and (3), and the result is as follows:

$$\begin{array}{ll}{B}_1& =A_1\circ R_1\\ & =\left(0.247,0.239,0.290,0.224\right)\circ \left[\begin{array}{cccc}0.1& 0.1& 0.2& 0.6\\ 0& 0.3& 0.3& 0.4\\ 0.1& 0.1& 0.4& 0.4\\ 0& 0.2& 0.3& 0.5\end{array}\right]\\ & =\left(0.054,0.170,0.304,0.472\right)\end{array}$$

Based on the maximum affiliation method, the maximum fuzzy evaluation result is 0.472, indicating that the level of CPC1 is “Inappropriate”. Similarly, the levels for all CPCs were determined, and the results are shown in

Table 8. The level of all CPCs associated with using target IVIS.

Factors	Definition	Level
CPC1	Degree of visualization of the interface	Inappropriate
CPC2	Rationalization of interaction mechanisms	Tolerable
CPC3	Adequacy of information feedback	Inefficient
CPC4	Number of simultaneous tasks	Inappropriate
CPC5	The cognitive load of drivers	High
CPC6	Personal health status	Healthy
CPC7	Available task execution time	Limited
CPC8	Cockpit environment of automobiles	Appropriate
CPC9	Driving skills and experience	Enough

.

4.3. Calculation of HIF and CPC Weights Related to IVIS

4.3.1. The Factor Centrality Weights of HIFs

This study integrated physiological and psychological dimensions to identify nine HIFs that affect drivers’ task performance [57,61]. The physiological factors identified were arm co-ordination (HIF1), fatigue resistance (HIF2), and attention level (HIF3). The psychological factors included emotional condition (HIF4), psychological tolerance (HIF5), memory ability (HIF6), reaction ability (HIF7), judgment ability (HIF8), and perceptual ability (HIF9).

The DEMATEL method was applied to calculate the factor centrality weights of HIFs for drivers interacting with the IVIS. First, the arithmetic mean of the member evaluations was calculated to derive the HIF direct relationship matrix X, and the results are presented in

Table 9. The HIF direct relationship matrix X.

Factors	HIF1	HIF2	HIF3	HIF4	HIF5	HIF6	HIF7	HIF8	HIF9
HIF1	0.000	1.571	1.714	1.571	1.286	1.714	1.429	1.571	1.714
HIF2	1.571	0.000	2.000	2.714	2.857	0.571	1.857	2.000	2.143
HIF3	1.000	1.286	0.000	2.429	1.571	2.000	2.000	1.857	2.143
HIF4	0.714	2.143	1.571	0.000	2.000	1.286	1.286	2.000	1.429
HIF5	0.714	1.714	1.143	2.286	0.000	1.714	1.286	1.714	1.857
HIF6	1.857	0.429	0.714	1.143	2.000	0.000	1.286	1.857	1.857
HIF7	2.714	2.000	2.571	2.143	1.286	1.571	0.000	2.000	2.143
HIF8	2.000	1.286	2.000	1.429	1.429	2.000	2.000	0.000	2.429
HIF9	1.429	1.714	2.143	2.429	1.714	1.857	2.143	1.000	0.000

. Then, the normalized direct relationship matrix D was determined using Equations (4) and (5). Finally, the total impact relationship matrix T was calculated using Equation (6), and the result is as follows:

$$T=\left[\begin{array}{ccccccccc}0.393& 0.494& 0.549& 0.610& 0.535& 0.517& 0.519& 0.544& 0.601\\ 0.566& 0.508& 0.672& 0.793& 0.721& 0.557& 0.642& 0.672& 0.740\\ 0.500& 0.531& 0.511& 0.717& 0.606& 0.583& 0.601& 0.614& 0.684\\ 0.434& 0.523& 0.542& 0.525& 0.573& 0.493& 0.511& 0.566& 0.587\\ 0.430& 0.498& 0.514& 0.640& 0.459& 0.510& 0.505& 0.546& 0.602\\ 0.450& 0.388& 0.445& 0.525& 0.514& 0.377& 0.459& 0.503& 0.549\\ 0.647& 0.631& 0.722& 0.786& 0.664& 0.628& 0.562& 0.693& 0.765\\ 0.561& 0.538& 0.631& 0.679& 0.607& 0.594& 0.611& 0.521& 0.710\\ 0.524& 0.558& 0.631& 0.725& 0.619& 0.579& 0.612& 0.578& 0.573\end{array}\right]$$

According to Equations (7) and (8), h_j and g_j can be obtained. By calculating h_j + g_j, the weight value v_j of HIFs can be derived according to Equation (10), and the results are presented in

Table 10. DEMATEL analysis results for HIFs.

Codes	Definitions	hj	gj	hj + gj	hj − gj	vj
HIF1	arm co-ordination	4.762	4.505	9.267	0.257	0.099
HIF2	fatigue resistance	5.870	4.667	10.537	1.203	0.113
HIF3	attention level	5.347	5.216	10.563	0.131	0.113
HIF4	emotional condition	4.753	5.999	10.752	−1.246	0.115
HIF5	psychological tolerance	4.705	5.298	10.003	−0.593	0.107
HIF6	memory ability	4.209	4.838	9.048	−0.629	0.097
HIF7	reaction ability	6.097	5.021	11.118	1.076	0.119
HIF8	judgment ability	5.452	5.236	10.688	0.216	0.115
HIF9	perceptual ability	5.399	5.812	11.211	−0.413	0.120

.

4.3.2. The Factor Centrality Weights of CPCs

This study used DEMATEL to evaluate the factor centrality weights of CPCs. After assessing the interactions among CPCs, the team calculated the CPC direct relationship matrix X′ using the arithmetic mean of the evaluation results. The results are presented in

Table 11. The CPC direct relationship matrix X′.

Factors	CPC1	CPC2	CPC3	CPC4	CPC5	CPC6	CPC7	CPC8	CPC9
CPC1	0.000	1.857	2.429	0.571	2.571	1.143	2.143	0.857	0.143
CPC2	1.286	0.000	2.714	2.286	2.571	0.429	2.286	0.000	1.429
CPC3	1.429	2.000	0.000	2.714	2.714	0.714	2.571	0.571	2.000
CPC4	1.000	1.857	2.286	0.000	2.857	1.429	2.857	0.857	2.000
CPC5	2.286	2.714	1.429	2.286	0.000	2.571	2.000	0.857	1.571
CPC6	2.000	0.857	0.714	2.143	3.000	0.000	1.429	0.571	2.286
CPC7	0.571	1.143	1.857	3.000	3.000	1.143	0.000	0.857	1.143
CPC8	1.429	0.000	1.143	1.143	2.000	0.286	1.429	0.000	1.286
CPC9	0.429	1.714	1.000	1.429	2.286	1.429	2.000	0.571	0.000

. Based on X′, the normalized direct relationship matrix D′ was computed using Equations (4) and (5). Finally, the total impact relationship matrix T′ was derived using Equation (6), and the result is presented as follows:

$$T^{\prime }=\left[\begin{array}{ccccccccc}0.398& 0.605& 0.650& 0.642& 0.883& 0.451& 0.737& 0.256& 0.460\\ 0.513& 0.566& 0.728& 0.805& 0.973& 0.463& 0.825& 0.235& 0.586\\ 0.561& 0.728& 0.631& 0.886& 1.059& 0.520& 0.903& 0.288& 0.662\\ 0.551& 0.729& 0.766& 0.754& 1.084& 0.566& 0.929& 0.308& 0.676\\ 0.631& 0.783& 0.739& 0.889& 0.949& 0.632& 0.901& 0.312& 0.662\\ 0.540& 0.595& 0.596& 0.761& 0.964& 0.424& 0.750& 0.260& 0.611\\ 0.468& 0.615& 0.663& 0.821& 0.972& 0.493& 0.676& 0.277& 0.563\\ 0.373& 0.373& 0.442& 0.505& 0.657& 0.307& 0.537& 0.155& 0.402\\ 0.394& 0.56& 0.533& 0.643& 0.815& 0.443& 0.684& 0.225& 0.418\end{array}\right]$$

The sum of rows h_i and the sum of columns g_i were computed using Equations (7) and (8), and the results are presented in rows 2 and 3 of

Table 12. DEMATEL analysis results for CPCs.

Factors	CPC1	CPC2	CPC3	CPC4	CPC5	CPC6	CPC7	CPC8	CPC9
hi	5.082	5.695	6.239	6.363	6.498	5.503	5.548	3.75	4.714
gi	4.43	5.553	5.748	6.706	8.356	4.299	6.942	2.318	5.04
hi + gi	9.512	11.248	11.987	13.069	14.854	9.802	12.49	6.068	9.754
hi − gi	0.651	0.142	0.491	−0.344	−1.858	1.204	−1.394	1.432	−0.326
wi	0.096	0.114	0.121	0.132	0.150	0.099	0.126	0.061	0.099

. Then, the factor centrality weights of CPCs were calculated using Equation (9), and the results are shown in row 6 of the table.

Based on the values of h_i and g_i in Table 12, the relationship between the degree to which CPCs influence and are influenced is illustrated in Figure 4a. The relationship between the centrality and the causality degree of CPCs is derived from h_i + g_i and h_i − g_i, and the results are shown in Figure 4b. The value of h_j − g_j reflects the degree of causality, with a positive value suggesting a cause factor and a negative value indicating an effect factor. According to Figure 4, the influence and centrality of CPC2, CPC3, CPC4, CPC5, and CPC7 are higher than the average values, indicating the need for correlation adjustment.

A visually simplified relationship matrix was developed to intuitively represent the interrelationships among CPCs. The threshold was set at the mean value of the full relational matrix, which can ensure a balanced cutoff, retaining significant inter-factor influences and filtering out negligible ones, thereby improving the stability and interpretability of the results. Only entries with t_ij greater than the threshold (i.e., the mean value of 0.6089) were retained to highlight the dominant interconnections among CPCs, thereby generating the simplified matrix T″, as shown in

Table 13. The CPC simplified relation matrix T″.

Factors	CPC1	CPC2	CPC3	CPC4	CPC5	CPC6	CPC7	CPC8	CPC9
CPC1	0.000	0.000	0.650	0.642	0.883	0.000	0.737	0.000	0.000
CPC2	0.000	0.000	0.728	0.805	0.973	0.000	0.825	0.000	0.000
CPC3	0.000	0.728	0.631	0.886	1.059	0.000	0.903	0.000	0.662
CPC4	0.000	0.729	0.766	0.754	1.084	0.000	0.929	0.000	0.676
CPC5	0.631	0.783	0.739	0.889	0.949	0.632	0.901	0.000	0.662
CPC6	0.000	0.000	0.000	0.761	0.964	0.000	0.750	0.000	0.611
CPC7	0.000	0.615	0.663	0.821	0.972	0.000	0.676	0.000	0.000
CPC8	0.000	0.000	0.000	0.000	0.657	0.000	0.000	0.000	0.000
CPC9	0.000	0.000	0.000	0.643	0.815	0.000	0.684	0.000	0.000

. The interactions between CPCs were visualized using the simplified matrix T″. The result is illustrated in Figure 5, where single arrows indicate unidirectional effects and double arrows indicate mutual effects.

4.3.3. The Cognitive Impact Weights of CPCs

The CREAM proposes that CPC levels influence cognitive function failures differently and that interactions among CPCs can modify these effects. This paper determined the effects and cognitive impact weights of cognitive function failures at each CPC level based on the traditional CREAM [33], and the results are shown in

Table 14. The effects and cognitive impact weights of CPCs on cognitive function failures.

Code	Level	Effect	Impact Weights on Cognitive Function Failures
			Observation	Interpretation	Planning	Execution
CPC1	Supportive	Improved	0.5	1.0	0.8	0.8
	Adequate	Not significant	1.0	1.0	1.0	1.0
	Tolerable	Not significant	1.0	1.0	1.0	1.0
	Inappropriate	Reduced	5.0	1.0	2.0	2.0
CPC2	Very sensible	Improved	0.8	1.0	0.5	0.8
	Reasonable	Not significant	1.0	1.0	1.0	1.0
	Tolerable	Not significant	1.0	1.0	1.0	1.0
	Unreasonable	Reduced	2.0	1.0	5.0	2.0
CPC3	Very efficient	Improved	1.0	0.8	1.0	0.8
	Efficient	Not significant	1.0	1.0	1.0	1.0
	Inefficient	Reduced	1.0	1.2	1.0	1.2
	Deficient	Reduced	1.0	2.0	1.0	2.0
CPC4	Appropriate	Improved	0.8	0.8	0.5	0.8
	Acceptable	Not significant	1.0	1.0	1.0	1.0
	Inappropriate	Reduced	2.0	2.0	5.0	2.0
CPC5	Low	Improved	1.0	0.8	1.0	0.8
	Medium	Not significant	1.0	1.0	1.0	1.0
	High	Reduced	1.0	2.0	1.0	2.0
CPC6	Healthy	Improved	0.5	1.0	1.0	0.8
	Generic	Not significant	1.0	1.0	1.0	1.0
	Unhealthy	Reduced	5.0	1.0	1.0	2.0
CPC7	Adequate	Improved	0.8	0.5	0.8	0.5
	Limited	Not significant	1.0	1.0	1.0	1.0
	Inadequate	Reduced	2.0	5.0	2.0	5.0
CPC8	Appropriate	Improved	0.5	1.0	1.0	0.8
	Acceptable	Not significant	1.0	1.0	1.0	1.0
	Inappropriate	Reduced	2.0	1.0	1.0	2.0
CPC9	Enough	Improved	0.5	0.5	0.5	0.8
	Limited	Not significant	1.0	1.0	1.0	1.0
	Insufficient	Reduced	5.0	5.0	5.0	2.0

.

According to the effect adjustment rules of the CREAM, CPC2, CPC3, CPC4, CPC5, and CPC7 were adjusted, and the results are presented in

Table 15. The adjustment rules and adjusted effect of CPCs.

CPC	Depends on the Following CPCs	Minimum Number of Synergetic CPCs	Original Effect	Adjusted Effect
CPC2	CPC3, CPC4, CPC5, CPC7	3 out of 4	Not significant	Reduced
CPC3	CPC1, CPC2, CPC4, CPC5, CPC7	4 out of 5	Reduced	Reduced
CPC4	CPC1, CPC2, CPC3, CPC5, CPC6, CPC7, CPC9	6 out of 7	Reduced	Reduced
CPC5	CPC1, CPC2, CPC3, CPC4, CPC6, CPC7, CPC8, CPC9	7 out of 8	Reduced	Reduced
CPC7	CPC1, CPC2, CPC3, CPC4, CPC5, CPC6, CPC9	6 out of 7	Not significant	Not significant

. For instance, if three or more factors in the four dependent CPCs of CPC2 show consistent improvements or reductions in their original effects, CPC2 is adjusted accordingly. The cognitive impact weights of CPCs on cognitive function failures were derived through a combined analysis, and the results are shown in

Table 16. The cognitive impact weights of CPCs on cognitive function failures.

Subtasks	CPC1	CPC2	CPC3	CPC4	CPC5	CPC6	CPC7	CPC8	CPC9
1.1	5.0	2.0	1.0	2.0	1.0	0.5	1.0	0.5	0.5
1.2.1	1.0	1.0	1.2	2.0	1.0	1.0	1.0	1.0	0.5
1.2.2	2.0	5.0	1.0	5.0	1.0	1.0	1.0	1.0	0.5
1.2.3	2.0	2.0	1.2	2.0	2.0	0.8	1.0	0.8	0.8
1.3.1	2.0	2.0	1.2	2.0	2.0	0.8	1.0	0.8	0.8
1.3.2	5.0	2.0	1.0	2.0	1.0	0.5	1.0	0.5	0.5
1.3.3.1	1.0	1.0	1.2	2.0	1.0	1.0	1.0	1.0	0.5
1.3.3.2	2.0	5.0	1.0	5.0	1.0	1.0	1.0	1.0	0.5
1.3.4	5.0	2.0	1.0	2.0	1.0	0.5	1.0	0.5	0.5
1.4.1	2.0	2.0	1.2	2.0	2.0	0.8	1.0	0.8	0.8
1.4.2	5.0	2.0	1.0	2.0	1.0	0.5	1.0	0.5	0.5

.

4.4. Calculation of the Cognitive Failure Probability Related to IVIS

The factor centrality weights of HIFs and CPCs were obtained based on the above analysis. The weight v_i for HIFs was obtained from Table 10. The weight w_i for CPCs was obtained from Table 12. The cognitive impact weight u_is for CPCs was obtained from Table 16. For subtask 1.2.2, the comprehensive weight was calculated using Equation (11), with the corresponding result listed in the last column of

Table 17. The integration of weights for subtask 1.2.2.

Cognitive Impact Weights for CPCs (ui)	Weights for CPCs (wi)	Weights for HIFs (vj)	Comprehensive Weights (pi)
2	0.096	0.099	1.939
5	0.114	0.113	5.044
1	0.121	0.113	1.071
5	0.132	0.115	5.739
1	0.15	0.107	1.402
1	0.099	0.097	1.021
1	0.126	0.119	1.059
1	0.061	0.115	0.530
0.5	0.099	0.12	0.413

, yielding a combined weight of 19.568. Then, using Equation (12), the CFP was calculated as 0.196 (0.010 × 19.568 ≈ 0.196). Similarly, the CFPs for other subtasks can be obtained, as shown in

Table 18. The CFPs of the drivers using the in-vehicle navigation function.

Subtasks	Nominal CFPs	Combined Weights	CFPs
1.2.2 Select a frequently used destination	0.010	19.568	0.196
1.3.3.2 Add waypoints	0.010	19.568	0.196
1.1 Click on the navigation icon	0.070	1.957	0.137
1.3.2 View navigation information	0.070	1.957	0.137
1.3.4 Return and continue navigation	0.070	1.957	0.137
1.4.2 Click the home icon	0.070	1.957	0.137
1.2.3 Choose a navigation route	0.003	7.695	0.023
1.4.1 Click the “Exit” button	0.003	7.695	0.023
1.2.1 Enter the destination name	0.010	0.939	0.009
1.3.3.1 Change the destination	0.010	0.939	0.009
1.3.1 Click the “Start” button	0.0005	7.695	0.004

Note: The bold underlined numbers represent the subtasks selected for corrective actions.

.

4.5. Recommendations for Design Improvements Related to IVIS

In interaction design, design improvements can lead to new usability issues [62]; therefore, the most critical problems must be prioritized for identification and resolution. The subtasks with the top 50% of CFPs were analyzed in depth (see Table 17). Task 1.2.2 (Select a frequently used destination) and task 1.3.3.2 (Add waypoints) exhibited the highest CFPs, with cognitive function failures P1 (Priority error) and P2 (Inadequate plan) occurring. Next, task 1.1 (Click on the navigation icon), task 1.3.2 (View navigation information), 1.3.4 (Return and continue navigation), and 1.4.2 (Click the home icon) showed cognitive function failures O2 (Wrong identification) and O3 (Observation not made).

To explore the causes of human error, the CREAM cause–effect matrix created by Hollnagel was employed to analyze the types of cognitive function failures and match the corresponding effects with causes [33]. A cause analysis was conducted based on the physiological and psychological characteristics of drivers, followed by the proposal of improvement measures [58,63]. First, for Priority error and Inadequate plan among the plan functions, the causes of human error included legitimate higher priorities, inattention, communication failures, uncertainty, distraction, errors in goals, design failures, parallel tasks, and excessive demands. Corresponding design improvements included optimizing the interface hierarchy, reducing distractions from secondary information, and providing timely interactive feedback (corrective actions for cognitive function failures P1 and P2). Then, for Wrong identification and Missed observation among the observation functions, the causes of human error included erroneous information, ambiguous symbol sets, missing information, ambiguous signals, information overload, hidden information, and multiple signals. Design improvements were focused on optimizing the interface layout, enhancing the visual style of the interface, and rationalizing the simultaneous presentation of information (corrective actions for cognitive function failures O2 and O3).

5. Discussion

This paper introduces a hybrid approach that combines the CREAM, FCE, and DEMATEL for human reliability analysis, focusing on cognitive demands in interaction design. By integrating FCE with the traditional CREAM, the subjectivity of expert evaluations in assessing CPC levels was reduced, enhancing the accuracy of the assessment. Additionally, the impact of HIFs and CPCs on cognitive function failures was comprehensively considered. To quantify the factor centrality weights of HIFs and CPCs, DEMATEL was employed, allowing for the analysis of the correlations between CPCs. Furthermore, to provide valuable insights for optimizing interaction design, the underlying causes of human error were explored. A case study of an IVIS design was conducted to demonstrate the proposed approach.

Based on the analysis of drivers using the in-vehicle navigation functions of the original IVIS, cognitive errors can be categorized into misinterpretation of functional information and misperception of visual symbols. The physiological and psychological characteristics of drivers contribute to these errors, with causes such as cognitive biases and workload leading to prioritization errors and misinterpretation during interaction. Additionally, failures in IVIS interface design and excessive task demands result in information overload and low visibility. Design improvements are proposed to mitigate human error in the IVIS.

The main page and the navigation page of the revised IVIS are presented in Figure 6. The research team reanalyzed the revised IVIS, and post-improvement CFPs were reassessed using the same method as the pre-improvement CFPs. To compare the CFPs before and after the improvement, A/B testing was employed. As a widely used method in human–computer interaction research, A/B testing allows for evaluating two versions of a design to identify the statistically more effective one [64]. The results showed that CFPs after improvement (M = 0.002, SD = 0.002) were significantly lower than CFPs before improvement (M = 0.092, SD = 0.078), t(10) = 3.79, p = 0.004. The box plot is shown in Figure 7, and the CFPs before and after improvement are marked using two * symbols, indicating a significant difference between the two groups. These results suggest a significant difference in CFPs during task completion with the revised IVIS, demonstrating the effectiveness of the proposed improvements and confirming the human reliability of the improved interaction system, thus validating the proposed methodology.

To further elaborate on the design improvement measures and their correlation with CPCs and HIFs, four optimization strategies were implemented to achieve a significant reduction in CFP, with their specific mapping relationships clarified as follows: (1) UI layout optimization, which effectively improved the visual attention allocation of CPCs; (2) feedback timing optimization, which reduced the response delay of HIFs; (3) icon redesign, which alleviated the cognitive ambiguity of HIFs; (4) hierarchy adjustment, which enhanced the information processing efficiency of CPCs. In addition, the reliability of the post-improvement CFP values was verified by two complementary methods: recalculation based on updated CPCs assessments, and repeated user experiments involving 30 subjects, which further confirmed the robustness of the improvement effects.

In order to consider the impact of CPC correlations on cognitive function failures, the traditional CREAM proposes basic adjustment rules. However, in the practical application of human reliability analysis for interactive systems, CPCs should be adjusted based on their interaction. To analyze the interrelationships among CPCs, the proposed method used DEMATEL. In the IVIS case study, the effects of CPCs were adjusted for interrelationships and the adjustment rules. For example, the original effect of CPC2 was not significant, and the adjusted effect was reduced. The proposed method provides a more accurate understanding of the effects of CPCs on cognitive function failures.

HIFs play a crucial role in user-centered interaction design, especially among IVIS users. However, the traditional CREAM does not focus on the impact of HIFs on human error in interactive systems during human reliability analysis. The proposed method identifies nine HIFs relevant to interactive systems operations and uses DEMATEL to examine their influence on human error. It also integrates the combined weights of CPCs and HIFs when calculating the CFPs. As a result, this approach offers more accurate outcomes, improving effectiveness and adaptability in practice.

To clarify the differences between the proposed method and related methods (including the traditional CREAM, single FCE, and single DEMATEL) in terms of output results and uncertainty handling, the details are presented in

Table 19. Comparison between related methods and the proposed method.

Method	Output Results	Uncertainty Handling
Traditional CREAM	CFP; no CPC weights or causal dependencies	No systematic method; ignores factor interaction uncertainty
Single FCE	Factor-level evaluation; simple weights; no CPC/causal info	Fuzzy membership functions; ignores incomplete factor uncertainty
Single DEMATEL	Factor causal dependencies; no CPC weights or CFP	Expert pairwise comparison; cannot handle CFP uncertainty
Proposed method	CPC weights, CFP, causal dependencies between CPCs and HIFs	Integrates FCE, DEMATEL, and CREAM to reduce multi-source uncertainty

. Compared with these related methods, the proposed method has the following innovation points: it constructs a human–computer interaction-specific CPC/HIF system, adopts DEMATEL for causal structure analysis of factors, uses FCE for quantitative assessment of factor levels, and integrates these three methods through a normalized product integration formula to form a systematic cognitive reliability analysis pipeline for interaction design.

This study was primarily conducted with drivers as the representative sample. Future research could include a broader range of participants to improve the applicability of the results. To resolve the limitations of the traditional CREAM in analyzing the interrelationships among CPCs, this study used DEMATEL to explore the interaction effects within interactive systems. This paper focused on CPC analysis and excluded the analysis of interaction among HIFs. Future research will investigate the mutual influences of HIFs to enhance human reliability analysis. Furthermore, this study mainly focused on identifying human errors and the causes of cognitive function failures. The interrelationship between cognitive function failures has not been explored in depth. Bayesian network analysis can be employed to explore the interrelationships among factors [26,37]. Bayesian network analysis effectively handles conditional dependencies, providing more accurate insights for optimizing human–computer interaction [65,66]. To comprehensively analyze human error in interactive systems, future research could integrate Bayesian networks with the proposed method to analyze the interrelationships among CPCs and the relationships between cognitive function failures. To ensure participant safety, the tests were conducted in static scenarios; however, future studies are planned for dynamic scenarios.

The proposed approach is relatively complex and involves multiple computational steps. Nevertheless, it is implemented systematically. More importantly, the approach integrates five well-defined simplification steps: analysis of cognitive function failures, assessment of CPC levels, calculation of HIF and CPC weights, calculation of the CFP, and recommendations for design improvements. With these detailed and standardized steps, the proposed approach is sufficiently reproducible. Although an IVIS was used as a case study, the approach is generalizable to other interactive systems for human error reduction. Future work will further validate and extend the approach by incorporating additional application cases.

6. Conclusions

This paper presents an approach for human reliability analysis in interactive systems design based on the CREAM, FCE, and DEMATEL. The approach employs the CREAM to analyze the cognitive demands of users in interactive systems. FCE is used to enhance the objectivity of expert evaluations of CPCs. DEMATEL is applied to explore the factor centrality weights of CPCs and HIFs on cognitive function failures and to analyze the interactions among CPCs. The cognitive reliability analysis was conducted by combining the cognitive impact weights of CPCs. Based on the analysis of human errors, design improvements are proposed. This approach provides a comprehensive framework for human reliability analysis in interaction design. Practical validation shows that the proposed approach yields a statistically significant reduction in the cognitive failure probability of the target interactive system, thereby fully confirming its effectiveness in improving human reliability.

This paper focused on user and system design aspects. Future research will aim to expand the scope to include a more integrated human reliability analysis, considering all components contributing to human error. Additionally, the following key directions should be emphasized in future studies. First, researchers should recruit a more diverse group of subjects to increase the sample size and improve the generalizability of the results. Second, users’ physical and psychological factors should be integrated into the CPC-HIF evaluation system to optimize the model. Third, a real-time online optimization system should be developed for dynamic human–computer interaction design, further reducing cognitive failure probability.