Research on the Design of a Basketball Shooting Training Machine Oriented Toward User Continuance Intention

Abstract

With limited coaching resources, automated training devices offer opportunities for self-directed sports practice. However, their practical value depends on users’ continued use. To identify the key determinants of continuance intention toward basketball shooting training machines, this study integrates the Unified Theory of Acceptance and Use of Technology (UTAUT) and Task–Technology Fit (TTF) into an analytical framework. A mixed-method design was adopted, including prototype experience, interviews, and questionnaire surveys. A total of 429 valid questionnaires were collected from basketball enthusiasts recruited from universities, fitness centers, and public basketball courts. The results indicate that performance expectancy, task–technology fit, and effort expectancy all positively influence continuance intention. Among these factors, performance expectancy shows the strongest direct effect (β = 0.44, p < 0.001). In addition, task–technology fit reinforces both performance expectancy and effort expectancy. To translate these findings into design practice, the study further integrates the Function Analysis System Technique (FAST) and the Function–Behavior–Structure (FBS) framework, generating a design pathway from behavioral mechanisms to functional elements and structural implementation. These findings provide theoretical and practical support for the design of automated training devices.

1. Introduction

China has made national fitness a strategic priority to advance public health and social development [1]. Basketball is one of the world’s most widely practiced sports, with more than 610 million participants. Shooting training is highly repetitive and readily quantifiable. It can produce measurable short-term improvement, enhance users’ sense of achievement, and increase sports participation [2]. At the same time, basketball skill acquisition requires extensive practice, and stable technical performance usually depends on long-term, structured, and repetitive training [3]. Expanding access to basic sports skill training may therefore help increase participation in physical activity at the population level.

Against the backdrop of the convergence of intelligent manufacturing and sports science, automated training systems integrating sensing, control, and feedback functions are reshaping sports training models [4]. Olsen R. J. and colleagues have shown that sports systems are rapidly evolving from simple performance monitoring toward broader applications, including rehabilitation and sports training [5]. As a result, data-driven training devices are becoming a practical approach to skill development. By collecting user data in real time through sensors, these systems shift skill learning from experience-based practice to data-driven training, providing a feasible path toward more scientific and accessible sports instruction [6]. However, the usage of public automated training devices commonly declines over time [7]. Once users’ intention to continue using such devices weakens, training effectiveness decreases and resource waste may occur [8]. Understanding continuance intention toward basketball shooting training machines has therefore become an important research issue.

User behavior theories provide a useful perspective for addressing this issue. The Unified Theory of Acceptance and Use of Technology (UTAUT) integrates several major user behavior models and is widely used to explain the acceptance and use of new technologies. Because it is derived from multiple theoretical foundations, it has shown strong predictive power across different contexts and offers good extensibility [9]. However, for highly task-oriented devices such as basketball shooting training machines, technology acceptance variables alone are insufficient to explain whether device functions truly fit training tasks. Empirical evidence from Hu S. and colleagues suggests that an integrated UTAUT–TTF model has stronger explanatory power than a single acceptance model in complex technology-use contexts, especially when task requirements and technological capabilities jointly shape user intention and continuance intention [10]. Basketball shooting training requires not only technical functions such as automatic ball return, landing-point control, and feedback analysis, but also effective alignment between these functions and users’ actual training goals. If the device fails to support training tasks effectively, users are unlikely to develop stable continuance intention, even if the interface is easy to use or the social evaluation is positive [11].

Task–Technology Fit (TTF) offers an important theoretical extension for this problem. TTF emphasizes that when the characteristics of a technology fit task requirements well, users are more likely to perceive stronger performance expectancy, that is, greater improvement in training outcomes. This in turn can strengthen their motivation for continued use. A higher level of fit may also indicate that device functions are more closely aligned with task demands, thereby reducing users’ cognitive burden and improving perceived ease of use, which further enhances effort expectancy [12]. For this reason, the present study incorporates TTF into UTAUT in order to improve the explanatory power of the model and provide a better explanation of continuance intention.

Existing studies on user behavior have mainly focused on testing relationships among behavioral variables, while relatively few have translated behavioral determinants into engineering-level design decisions. This gap is particularly evident in technology-rich usage contexts [13]. For automated training devices, the practical value of research remains limited if empirical findings cannot be converted into design strategies for real training scenarios. This reflects a persistent disconnect between user behavior research and engineering design practice.

Against this background, this study focuses on continuance intention toward basketball shooting training machines. It aims to identify the key factors affecting users’ continuance intention and to further explore how empirical findings can be translated into product design strategies.

To achieve this goal, the study integrates UTAUT and TTF to construct an explanatory model of continuance intention in the context of basketball shooting training machines. A prototype-based user study was first conducted to identify key needs related to training effectiveness, operational burden, and task fit. Based on these findings, measurement scales were developed and a questionnaire survey was carried out. Structural equation modeling was then used to test the relationships among the variables. In addition, the Function Analysis System Technique (FAST) and the Function–Behavior–Structure (FBS) framework were employed to translate key demand factors affecting continuance intention into functional elements and component configurations. On this basis, a research pathway was established: Problem Definition–Prototype-Based User Study–Needs Extraction and Construct Development–Questionnaire Design and Data Collection–SEM Analysis and Hypothesis Testing–Design Generation and Evaluation, as shown in Figure 1. The basketball shooting training machine prototype is shown in Figure 2.

2. Materials and Methods

2.1. Unified Theory of Acceptance and Use of Technology (UTAUT)

Venkatesh et al. proposed the Unified Theory of Acceptance and Use of Technology (UTAUT) in 2003 by integrating eight models, including the Technology Acceptance Model (TAM), the Theory of Planned Behavior (TPB), and Innovation Diffusion Theory [14]. UTAUT introduces four core constructs—performance expectancy, effort expectancy, social influence, and facilitating conditions—to provide a more comprehensive framework for explaining users’ behavioral intention toward technology use. Owing to its strong extensibility, UTAUT has shown greater explanatory power for usage intention than single-theory models.

In the field of sports research, UTAUT has been further validated and extended. Integrated models that combine UTAUT with other need- and perception-related theories have been increasingly adopted in sports and wearable technology studies. These studies indicate that effort expectancy, social influence, and facilitating conditions remain important determinants of continuance intention. For example, Liu et al. applied UTAUT to explain users’ intention to adopt physical monitoring products. In their study, performance expectancy was defined as users’ expectations of exercise-related benefits, while social influence was treated as the effect of group atmosphere. Their results confirmed the important roles of performance expectancy and social influence in shaping usage intention [15]. Yang M. also employed the UTAUT framework to examine users’ choice of intelligent sports products and found that usage decisions were highly dependent on social influence and facilitating conditions. This study provided useful support for improving system design processes [16].

However, for task-oriented training devices, acceptance-related factors alone are insufficient to explain how device functions support training tasks and ultimately affect training outcomes. Therefore, introducing the perspective of Task–Technology Fit (TTF) is expected to provide a more effective explanation of users’ actual needs.

2.2. Task–Technology Fit (TTF)

Task–Technology Fit (TTF) was proposed by Goodhue and Thompson [17]. This theory evaluates the effects of technology adoption by examining the consistency between task requirements and technological characteristics. It emphasizes that such fit affects both end-user performance and technology utilization. When users perceive that a system can significantly improve task efficiency, their expectations toward technology use are strengthened.

In research on continuance intention in information systems, Lin demonstrated that TTF enhances the effects of performance and continuance intention, and that an integrated model provides stronger explanatory power than traditional single-theory models [18]. Salhieh et al. combined TTF with UTAUT to explain the relationship between performance expectancy and the fit between technology and task requirements [19]. Liu et al. further noted that studies on sports equipment often overlook whether device functions genuinely match the requirements of sports activities [15].

In automated sports training, training goals can only be effectively achieved when functional configurations properly support task execution. In practical applications, technology implementation must therefore consider the coupling between task requirements and technological functions in order to realize essential functions [20]. For this reason, the present study combines UTAUT and TTF to explain user needs in complex technology-use scenarios.

2.3. Function Analysis System Technique (FAST)

The Function Analysis System Technique (FAST) organizes system functions through a “Why–How” logic. It identifies hierarchical and causal relationships among functions and emphasizes starting from functional objectives in the early design stage to decompose structures and construct a complete functional system. Zhao et al. demonstrated the feasibility of integrating FAST with decision-making methods and showed that this integration improved the scientific rigor and systematic nature of design decisions [21]. Hu et al. used FAST to transform weighted user needs into functional components and constructed a functional system model for equipment design, thereby demonstrating the effectiveness of FAST in complex system architectures [22]. In another study, Li et al. combined FAST, Kano, and Quality Function Deployment (QFD) to translate user needs into technical design solutions, further confirming the compatibility of FAST with user behavior theories [23].

By structuring functions and clarifying causal relationships among them, FAST systematically links behavior-related user needs with functional decomposition in conceptual design. In the context of automated sports training equipment, FAST enables abstract behavioral determinants to be translated into an operable functional framework.

2.4. Function–Behavior–Structure (FBS)

The Function–Behavior–Structure (FBS) model is a systematic design theory framework that decomposes complex design problems into three levels: function, behavior, and structure. The model emphasizes the dynamic mapping among these three levels. By analyzing appropriate functional constraints, it helps adapt design solutions to user behavior and determine feasible functional schemes [24]. Cascini et al. extended the FBS framework by incorporating user needs and contextual factors [25]. Through recursive representation across the function–behavior–structure hierarchy, FBS supports structural semantic analysis and design advancement in complex systems. Further studies have demonstrated the applicability of contextualized FBS in analyzing collaborative interactions between users and designers, reinforcing its value in representing complex behavior-based design processes.

Based on the functional hierarchy established by FAST, this study applies FBS to achieve a vertical transition from function to behavior and then to structure. The introduction of FBS avoids limiting functional decomposition to the conceptual level alone. Instead, it enables functional configurations to be translated into module design, control strategies, and interaction structures for automated sports training devices. In this way, the study extends behavioral explanation into engineering practice. The Research framework is shown in Figure 3.

2.5. Prototype-Based User Study

A prototype-based user study was conducted to ensure that the research hypotheses could, to some extent, reflect users’ needs in real training scenarios [26]. The study was based on a basketball shooting training machine project jointly developed by a university and a company in Wuhan. Since January 2024, the company has deployed experimental prototypes in several public spaces in Wuhan, including parks and residential communities, to test public acceptance. The prototype provides basic functions such as repetitive shooting practice and automatic ball return, using the Lefu Self-Service Basketball Shooting Training Machine (Wuhan Lefu Operation and Distribution Co., Ltd., Wuhan, China). Although its training modes are limited, it can still capture users’ real reactions in public-use settings.

The research team conducted the prototype-based user study in April 2025. A total of 18 participants with basketball-related experience were recruited through basketball enthusiast communities and on-site invitations. The inclusion criteria were as follows: participants have basic basketball experience, and voluntarily agree to participate in the study.

Each participant completed approximately 15–20 min of hands-on experience with the prototype, followed by a 10 min interview. During the experience session, participants interacted with the prototype in a simulated shooting-training scenario. The observation focused on training rhythm, perceived operational burden, feedback perception, and task fit. The follow-up interviews further explored participants’ expectations, difficulties, and suggestions regarding the training device. Among the participants, 12 were male (66.7%) and 6 were female (33.3%). In terms of age, 7 participants were aged 18–25 years (38.9%), 6 were aged 26–35 years (33.3%), and 5 were aged 36 years or above (27.8%). The 18–25 group represents emerging adulthood, the 26–35 group represents a more stable adult stage, and the 36+ group represents a more mature stage of adulthood. This grouping helps reflect differences across adult life stages in sports participation habits, device acceptance, and training needs [27]. The study procedure is shown in Figure 4.

All participants were enrolled in the study after providing informed consent.

Table 1. Initial User Interview Records.

Participant	Relevant Information	Original Interview Statement
P1	Male, 18, <1 year experience	I am not quite sure what each training mode is actually designed for. I hope the system can make it clear at a glance what each mode is suitable for.
P2	Male, 20, 1–3 years experience	The ball-feeding mode is too limited.
P3	Male, 21, <1 year experience	There is no safety protection, and I am worried about damaging the equipment.
P4	Male, 22, >3 years experience	There is a lack of movement or technique instruction.
P5	Male, 22, >3 years experience	The level of interactivity is insufficient; practicing alone feels boring.
P6	Male, 22, <1 year experience	Shooting machines are prone to damage and have a short service life. The ones I usually see in parks are often already broken.
P7	Male, 23, 1–3 years experience	I worry about damaging public equipment, so I do not feel comfortable using it freely.
P8	Male, 30, 1–3 years experience	Frequent use may lead to a high failure rate.
P9	Male, 32, 1–3 years experience	The target user group is too narrow, and there is a lack of a dedicated training system.
P10	Male, 35, >3 years experience	Product homogeneity is serious; most shooting machines do not seem very different from one another.
P11	Male, 36, >3 years experience	Many products on the market have unusual and outdated designs.
P12	Male, 41, >3 years experience	Sometimes the ball delivery rhythm and landing point do not match the content I want to practice.
P13	Female, 22, <1 year experience	The data provided after training are too basic and do not offer much reference value.
P14	Female, 27, <1 year experience	If the ball delivery is unstable or the landing point is inaccurate, the overall training quality will be affected.
P15	Female, 26, 1–3 years experience	My training goals may change over time, but the content provided by the device does not adapt accordingly.
P16	Female, 38, 1–3 years experience	The data currently provided are not sufficiently in-depth and cannot truly help me analyze my problems.
P17	Female, 28, 1–3 years experience	If the device cannot fit into my existing training routine, I will not continue using it in the long term.
P18	Female, 45, >3 years experience	Once the device malfunctions or responds slowly even once, I may lose the willingness to keep using it.

summarizes the participants’ demographic information and the initial interview statements collected during the prototype-based user study.

3. Model Construction and Research Assumptions

3.1. Model Construction

This study constructs an integrated theoretical model by combining UTAUT and TTF. The influencing factors specified in the model serve as independent variables, while continuance intention is treated as the dependent variable. The core constructs of UTAUT primarily explain the motivational mechanisms underlying users’ technology acceptance and use, whereas TTF emphasizes the degree of alignment between task demands and technological capabilities. Their integration enables the interpretation of use behavior from two complementary perspectives—user cognition and task–technology alignment—thereby providing a systematic theoretical foundation for understanding continuance intention toward public sport-training equipment. The resulting UTAUT–TTF framework (Figure 5) is designed principally for basketball enthusiasts engaged in self-directed training, and is used to explain their continuance intention to use a basketball shooting training machine in autonomous practice contexts.

3.2. Research Hypotheses

Based on the original interview data presented in Table 1, the research team identified and categorized user needs related to training effectiveness, operational convenience, and functional–technical fit. These needs were then linked to the constructs in the integrated UTAUT–TTF framework. The qualitative study helped identify context-specific user needs, provided a basis for hypothesis development and questionnaire design, and improved the alignment between the proposed hypotheses and real usage scenarios [28], as shown in Figure 6.

To explore the relationships among the factors influencing users’ continuance intention toward the basketball shooting training machine, the research team reviewed the relevant literature and developed the following hypotheses.

Performance expectancy refers to the degree to which users believe that the basketball shooting training machine can improve their training outcomes and basketball skills. For this type of device, its core value lies in enhancing training efficiency through automated ball delivery and data feedback [28]. When users perceive that the device can significantly improve training outcomes, they are more likely to recognize its practical value and develop stronger continuance intention. Accordingly, we propose Hypothesis (H):

H1: 

Performance Expectancy positively influences users’ continuance intention to use the basketball shooting training machine.

Effort expectancy refers to the degree to which users perceive the basketball shooting training machine as easy to learn and easy to operate. For basketball enthusiasts without a professional background, the usability of the human–machine interface is particularly important. When the device is easy to operate and provides a user-friendly interface, it can reduce barriers to use and lower the effort required for learning and operation. This, in turn, increases users’ willingness to use the device for basketball training tasks [29]. Based on this reasoning, the following hypothesis is proposed:

H2: 

Effort Expectancy positively influences users’ continuance intention to use the basketball shooting training machine.

Social influence refers to the extent to which other people’s opinions and recommendations affect an individual’s decision to use the basketball shooting training machine. Li et al. examined social influence from three dimensions—subjective norms, social image, and social identity—and found that these social factors significantly affect continuance intention through mechanisms such as perceived usefulness [30]. In sports contexts, users are usually embedded in social networks, where recommendations from peers and public evaluations can significantly shape their judgments of value. In the present study, social influence specifically refers to the effects of surrounding people’s evaluations, recommendations, and supportive attitudes toward the shooting training machine on users’ continuance intention. Based on this reasoning, the following hypothesis is proposed:

H3: 

Social Influence positively influences users’ continuance intention to use the basketball shooting training machine.

Facilitating conditions refer to the degree of external support available to users when using the device, encompassing maintenance services, technical guidance, and facility provisions. When users can obtain sufficient external support, barriers encountered during use are mitigated, thereby enhancing continuance intention, hence the following hypothesis:

H4: 

Facilitating Conditions positively influence users’ continuance intention to use the basketball shooting training machine.

Technology characteristics denote the functional attributes and service quality of the technological tool. In basketball training, relevant characteristics of a shooting training machine include adjustable feeding frequency, controllable landing points, and real-time performance feedback. When technical capabilities more precisely cover training requirements, task–technology fit is expected to improve. Accordingly:

H5: 

Technology Characteristics positively influence Task–Technology Fit.

Task characteristics refer to the specific requirements and degree of complexity that users encounter while performing a task. When these characteristics align closely with users’ needs, they can markedly strengthen users’ sense of control and autonomy. In other words, clearly defined task characteristics tend to increase users’ reliance on technological products and, in turn, reinforce user stickiness (i.e., sustained engagement and continued use). For basketball enthusiasts, training tasks typically encompass multiple goals, such as improving shooting accuracy, correcting shooting mechanics, and enhancing physical conditioning. As training objectives become more explicit—and as task difficulty and intensity increase—users become more sensitive to the extent to which an automated training device can match their functional requirements. Under such conditions, they are also more likely to perceive a high level of fit during practice. Accordingly:

H6: 

Task Characteristics positively influence Task–Technology Fit.

Within TTF theory, task–technology fit reflects the extent to which the training task aligns with the device’s technical capabilities. Technology characteristics and task characteristics are two critical antecedents of fit. When the machine’s technological attributes are highly congruent with users’ task requirements, users are more likely to obtain superior training experiences [20], which in turn strengthens continuance intention. Therefore:

H7: 

Task–Technology Fit positively influences users’ continuance intention to use the basketball shooting training machine.

TTF theory posits that users’ perceptions of technology use and their continuance intention are fundamentally contingent upon the degree of alignment between technological features and task demands. In the context of basketball shooting training, when the device’s functionalities closely match users’ needs for skill enhancement, users are more likely to clearly perceive improvements in their shooting accuracy and postural standardization, thereby substantially bolstering their performance expectancy toward the equipment. Within integrated UTAUT–TTF research frameworks, numerous studies have shown that task–technology fit is a key antecedent of performance expectancy, which depends heavily on the alignment between technology and users’ task requirements [31]. Furthermore, when the device’s functional design and operational logic align seamlessly with users’ training tasks, it significantly alleviates perceived operational burdens, thereby elevating effort expectancy. Accordingly, this study proposes the following hypotheses:

H8: 

Task–Technology Fit positively influences users’ Performance Expectancy toward the basketball shooting training machine.

H9: 

Task–Technology Fit positively influences users’ Effort Expectancy toward the basketball shooting training machine.

3.3. Questionnaire Design

The questionnaire consisted of two parts. The first part collected demographic information, including gender, age, education level, and basketball experience. The second part measured the constructs in the UTAUT and TTF frameworks, including performance expectancy, effort expectancy, social influence, facilitating conditions, task characteristics, technology characteristics, task–technology fit, and continuance intention. All items were designed to capture respondents’ actual perceptions. An example item is: “I believe the basketball shooting training machine can improve my shooting accuracy within a short period of time.”

Most items were adapted from established scales and revised according to findings from the prototype-based user study to ensure their suitability for the basketball shooting training context. All items were measured on a five-point Likert scale, ranging from 1 (“strongly disagree”) to 5 (“strongly agree”). Each latent construct was measured with two to three items, resulting in a total of 22 items.

Before the formal survey, a pilot test was conducted with 35 respondents who had basketball participation experience. All participants watched a short introduction to the basketball shooting training machine before completing the questionnaire. Based on the pilot feedback, several items were revised to improve clarity.

3.4. Data Collection and Measurement Items

To test the above hypotheses, the research team designed a survey instrument and conducted data collection. The questionnaire included respondents’ demographic information, as well as seven independent variables and one dependent variable specified in the research model. To ensure data validity, screening items were incorporated to exclude invalid responses. All measurement items were assessed using a five-point Likert scale, where 1 indicates “strongly disagree” and 5 indicates “strongly agree.” The full set of measurement items is presented in

Table 2. Description and Source of the Measurement Items.

Variable	Questionnaire Code	Content of the Questionnaire	Source
Performance Expectancy	PE1	I believe the basketball shooting training machine can improve my shooting accuracy within a short period of time.	[31]
	PE2	I believe the repetitive training function provided by the basketball shooting training machine is significantly helpful to my practice.
	PE3	The shooting training function of the basketball shooting training machine is important to me.
Effort Expectancy	EE1	I find the basketball shooting training machine easy to learn how to use.	[32]
	EE2	I find the functional interface of the basketball shooting training machine clear and easy to understand.
	EE3	I believe it requires little effort to learn how to operate the basketball shooting training machine.
Social Influence	SI1	I think it is appropriate to use the basketball shooting training machine as a supplementary training tool.	[11]
	SI2	I would recommend the use of the basketball shooting training machine to people around me.
	SI3	I would follow others’ recommendations to use the basketball shooting training machine for practice.
Facilitating Conditions	FC1	I believe the basketball shooting training machine is easy to maintain.	[11]
	FC2	When encountering technical problems while using the basketball shooting training machine, I am able to resolve them independently.
	FC3	When I encounter difficulties using the basketball shooting training machine, I am able to obtain assistance.
Task Characteristics	TAC1	I hope the basketball shooting training machine can help me identify bottlenecks in my training.	[33]
	TAC2	I need to engage in free-throw practice to improve my shooting skills.
	TAC3	I need the recording function of the basketball shooting training machine, as I believe it enables me to evaluate my training progress.
Technology Characteristics	TEC1	I believe the real-time recording function of the basketball shooting training machine would be very useful.
	TEC2	I believe the personalized training plan function of the basketball shooting training machine would be very useful.
	TEC3	I believe the training report feedback function of the basketball shooting training machine would be very useful.
Task–Technology Fit	TTF1	The feedback and recommendation function of the basketball shooting training machine aligns well with my training plan.
	TTF2	The functions of the basketball shooting training machine meet my training needs.
Continuance Intention	CI1	I intend to continue using the basketball shooting training machine for practice in the future.	[34]
	CI2	I will maintain my frequency of using the basketball shooting training machine.

.

3.5. Descriptive Statistical Analysis

The formal survey was conducted using a combination of online and offline methods. A total of 562 questionnaires were distributed. After data screening, 429 valid responses were retained, yielding an effective response rate of 76.3%. The online sample was mainly recruited through social media platforms and basketball enthusiast communities. The offline sample was collected from universities, fitness centers, and public basketball courts. All participants were informed of the research purpose before completing the questionnaire and participated voluntarily after providing informed consent.

To improve data quality, explicit screening criteria were applied. Responses were excluded if they contained missing information, showed obvious response patterns, had an abnormally short completion time, or failed the logic checks of the screening items. The final valid sample met the requirements for structural equation modeling and provided a reliable basis for subsequent model testing. According to the survey results, male participants accounted for 67.4% of the sample, while female participants accounted for 32.6%. The largest group being 18–25 years old (50.3%). Participants with a bachelor’s degree accounted for 62.2% of the sample. In addition, 31.9% of respondents had prior experience using automated training equipment. Detailed descriptive statistics of the sample are presented in

Table 3. Demographic Characteristics of the Sample.

Variable	Category	Frequency	Percent
Gender	Male	288	67.4
	Female	141	32.6
Age	under 18	13	3
	18–25	216	50.3
	26–35 years	138	32.2
	36 years and above	62	14.5
Education level	High School and below	96	22.4
	Bachelor	267	62.2
	Master and above	66	15.4
Use experience	Yes	137	31.9
	No	292	68.1

.

3.6. Confirmatory Factor Analysis (CFA) and Model Fit

To assess the reliability and validity of the questionnaire data, internal consistency reliability was first examined for each latent construct using SPSS 27.0. The results showed that all Cronbach’s alpha values were above 0.70, indicating good internal consistency of the measurement scales, as shown in

Table 4. Reliability verification analysis.

Measured Variables	Scale Items	Cronbach’s Alpha
PE	3	0.856
EE	3	0.904
SI	3	0.897
FC	3	0.840
TAC	3	0.911
TEC	3	0.829
TTF	2	0.894
CI	2	0.868

.

To examine sampling adequacy, the Kaiser–Meyer–Olkin (KMO) value and Bartlett’s test of sphericity were calculated. The KMO value was 0.898, and Bartlett’s test was significant (p < 0.001), indicating that the data were suitable for factor analysis. Exploratory factor analysis further confirmed the preliminary structure of the scale. The cumulative variance explained reached 73.792%, suggesting that the scale had a sound structural basis.

Based on the exploratory factor analysis, this study established a factor model of the determinants of users’ continuance intention toward the basketball shooting training machine. Confirmatory factor analysis was then conducted using AMOS 31.0 to test the structural validity of the model. The model fit indices met commonly accepted criteria, indicating an overall acceptable model fit. The detailed results are presented in

Table 5. Model fit analysis results.

Fitness Index	Metric	Criteria for Judgment	Fitting Situation
CMID	556.836	The smaller the better	Acceptable
DF	194	The smaller the better	Ideal
CMID/DF	2.870	<3 (Ideal); <5 (Acceptable)	Ideal
GFI	0.862	>0.8 (Acceptable); 0.9 (Ideal)	Acceptable
AGFI	0.844	>0.8 (Acceptable); 0.9 (Ideal)	Acceptable
CFI	0.947	>0.9	Acceptable
TLI (NNFI)	0.931	>0.8 (Acceptable); 0.9 (Ideal)	Acceptable
RMSEA	0.066	<0.1 (Acceptable); <0.08 (Ideal)	Acceptable

.

Convergent validity was further examined. The results showed that the standardized factor loadings of all measurement items were greater than 0.50 and statistically significant [27]. In addition, the composite reliability (CR) values of all latent constructs exceeded 0.70, and the average variance extracted (AVE) values were all above 0.50. These results indicate good convergent validity of the measurement scale. The detailed results are shown in

Table 6. Confirmatory Factor Analysis and Convergent Validity Test.

Variable	Questionnaire Code	Significance Estimation			Item Reliability		Composite Reliability	Convergent Validity
		Unstd.	S.E.	P	Std.	SMC	CR	AVE
CI	CI1	1.000			0.800	0.640	0.704	0.545
	CI2	0.736	0.067	***	0.671	0.450
PE	PE1	1.000			0.893	0.797	0.876	0.703
	PE2	1.161	0.078	***	0.873	0.762
	PE3	1.248	0.085	***	0.741	0.549
EE	EE1	1.000			0.922	0.850	0.881	0.713
	EE2	1.188	0.078	***	0.842	0.709
	EE3	1.205	0.076	***	0.762	0.581
SI	SI1	1.000			0.821	0.674	0.825	0.613
	SI2	1.242	0.104	***	0.806	0.650
	SI3	1.222	0.099	***	0.717	0.514
FC	FC1	1.000			0.863	0.745	0.801	0.576
	FC2	1.146	0.108	***	0.749	0.561
	FC3	1.234	0.113	***	0.650	0.423
TEC	TEC1	1.000			0.779	0.607	0.787	0.556
	TEC2	1.263	0.126	***	0.829	0.687
	TEC3	1.097	0.112	***	0.712	0.506
TAC	TAC1	1.000			0.639	0.408	0.828	0.621
	TAC2	0.937	0.063	***	0.822	0.676
	TAC3	0.780	0.069	***	0.882	0.778
TTF	TTF1	1.000			0.945	0.893	0.871	0.775
	TTF2	0.977	0.066	***	0.743	0.552

Note: *** indicates p < 0.001.

.

3.7. Path Analysis and Hypothesis Testing

This study applied an integrated UTAUT–TTF framework to investigate the factors influencing users’ continuance intention. The exogenous variables included Task–Technology Fit, Performance Expectancy, Effort Expectancy, Facilitating Conditions, and Social Influence. The path analysis results are shown in Figure 7. Performance Expectancy significantly increased continuance intention (β = 0.44, p < 0.001), supporting H1. Effort Expectancy also showed a significant positive effect (β = 0.24, p < 0.001), supporting H2. Social Influence (β = 0.19, p < 0.001) and Facilitating Conditions (β = 0.10, p < 0.05) were likewise significant, supporting H3 and H4. Technology Characteristics (β = 0.49, p < 0.001) and Task Characteristics (β = 0.27, p < 0.01) both significantly improved Task–Technology Fit, supporting H5 and H6. Task–Technology Fit further exerted a significant positive effect on continuance intention (β = 0.32, p < 0.001), supporting H7. In addition, Task–Technology Fit significantly enhanced Performance Expectancy (β = 0.61, p < 0.001) and Effort Expectancy (β = 0.57, p < 0.001), supporting H8 and H9. Overall, all nine hypotheses were supported, indicating that the integrated UTAUT–TTF model effectively explains users’ continuance intention toward the basketball shooting training machine. The hypothesis testing results are reported in

Table 7. Hypothesis Testing Results and Path Coefficients.

Serial	Hypothetical	Path Factor	P	Results
H1	PE → CI	0.44	***	Supported
H2	EE → CI	0.24	***	Supported
H3	SI → CI	0.19	***	Supported
H4	FC → CI	0.10	*	Supported
H5	TEC → TTF	0.49	***	Supported
H6	TAC → TTF	0.27	**	Supported
H7	TTF → CI	0.32	***	Supported
H8	TTF → PE	0.61	***	Supported
H9	TTF → EE	0.57	***	Supported

Note: *** indicates p < 0.001, ** indicates p < 0.01, * indicates p < 0.05.

.

4. Design Strategies and Case Studies

4.1. Normalization of Path Coefficients

Path coefficients output by the Structural Equation Model (SEM) reflect the strength of the marginal effect of each influencing factor on users’ continuance intention. To quantitatively clarify the relative importance of these factors, this study employs a normalization process to convert path coefficients into comparable and additive requirement weights, thereby enabling clear quantification of the relative significance of each determinant. The formula for calculating the normalized weights is as follows: $\begin{array}{c}{\mathrm{W}}_{\mathrm{i}}={\displaystyle \frac{{\mathsf{\beta }}_{\mathrm{i}}}{\sum _{\mathrm{j}=1}^{\mathrm{n}}{\mathsf{\beta }}_{\mathrm{j}}}}\end{array}$. In the formula, Wi denotes the normalized weight of the i-th factor, β_i represents the path coefficient of the i-th factor, and n is defined as the total number of factors directly influencing continuance intention (CI). In this study, the paths directly exerting an effect on continuance intention (CI) include: Performance Expectancy (PE → CI, β = 0.44), Effort Expectancy (EE → CI, β = 0.24), Task–Technology Fit (TTF → CI, β = 0.32), Social Influence (SI → CI, β = 0.19), and Facilitating Conditions (FC → CI, β = 0.10).

By substituting the aforementioned path coefficients into the formula, the normalized weights of each factor were calculated. Detailed results are presented in

Table 8. Results of Path Coefficient Normalization.

Variable	Hypothetical	Path Factor	Weight (%)
PE	PE → CI	0.44	34.11
TTF	TTF → CI	0.32	24.81
EE	EE → CI	0.24	18.60
SI	SI → CI	0.19	14.73
FC	FC → CI	0.10	7.75
Sum		1.29	100.00

.

4.2. FAST Function Decomposition

Based on the results of user requirement prioritization, this study employs the Function Analysis System Technique (FAST) to conduct a structured decomposition of user requirements. FAST is built on the hierarchical logic framework of why it is needed → what it entails → how it is implemented, which translates abstract user requirements into actionable, multi-level functional hierarchies. On this basis, we derived the requirement weight (W) via normalization of path coefficients from the Structural Equation Model (SEM), and we further classified the functional hierarchies to finalize the implementation priority of each function.

Core Driving Requirements (W ≥ 30%) correspond to Performance Expectancy (PE, W = 34.11%). This factor exerts the dominant predictive effect on user continuance intention, serving as the primary driver of long-term device usage.

Critical Supporting Requirements (10% < W < 30%) correspond to Effort Expectancy (EE, W = 18.30%), Social Influence (SI, W = 14.73%), and Task–Technology Fit (TTF, W = 24.81%). These factors deliver substantial explanatory power for continuance intention, acting as critical secondary drivers of sustained user engagement.

Optimization and Expansion Requirements (W ≤ 10%) correspond to Facilitating Conditions (FC, W = 7.75%). This factor exhibits a weaker marginal effect and secondary explanatory power, yet remains an indispensable component of the holistic user requirement framework.

Building on the above analysis, we prioritized functional implementation in alignment with the weight of each determinant. This hierarchical prioritization provides a quantitative, evidence-based decision-making framework for the mapping and translation from function → behavior → component in product design, as illustrated in Figure 8.

4.3. FBS Function–Structure Mapping

With functional priorities established, the next step is to translate the key elements of function realization into actionable structural solutions. To this end, we introduced the Function–Behavior–Structure (FBS) model to progressively map functional elements to structural elements, thereby providing component-level references for the design prototype, as illustrated in Figure 9.

• Function–Behaviour Mapping:

The function–behaviour mapping follows the logic of “functional intent → behavioural manifestation.” In particular, the function Recognise Shooting Posture is mapped to the behaviour Detect Body Motion State. The function Track Motion Trajectory is mapped to behaviours including Localise Ball Position and Collect Basketballs, enabling localization of ball movement paths and repeated ball collection. The function Record Training Results is mapped to Log Scored Shots, while Monitor Training Trends is mapped to Generate Performance Analytics.

For device operation, Simplify User Interaction is mapped to behaviours such as Execute One-Touch Activation and Deliver User Instructions, thereby reducing operational burden. The function Simulate Game Scenarios is mapped to training behaviours including Enable Multi-Angle Ball Launch and Regulate Ball Release Rhythm, enhancing ecological validity and realism of training contexts. Visualise Shooting Deviation is mapped to Display Shot Deviation, and Provide Equipment Protection corresponds to Absorb Impact Forces, ensuring training safety.

It should be noted that the functions derived from FAST already exhibit clear behavioural orientations. Therefore, during FBS reasoning, certain functions also served directly as sources for behaviour mapping. In addition, extended functions such as social engagement features and points-based rewards are treated as auxiliary services and are not elaborated in the structural design of the device body.

• Behaviour–Structure Mapping:

The behaviour–structure mapping is guided by the logic of “behaviour realization → physical carrier.” For data acquisition and analysis behaviours, Detect Body Motion State corresponds to the Camera Module. Localise Ball Position and Collect Basketballs correspond to a Positioning Marker and a Ball Collection Net, respectively. The behaviour Log Scored Shots corresponds to the Processing Unit and a Scoreboard.

For device operation, Execute One-Touch Activation corresponds to a One-Touch Start Button and a QR Code Scanner, integrating both physical-button control and mobile mini-program control. Training behaviours—namely Enable Multi-Angle Ball Launch and Regulate Ball Release Rhythm—correspond to a Ball Launching Mechanism, a Ball Storage Unit, and a Rotating Base Platform for controlling launch direction, complemented by a Launch Direction Indicator and a power module. The behaviour Guide User Interaction corresponds to a Speaker Module and a Guidance Light, enabling multimodal (audio–visual) guidance. The behaviour Display Shot Deviation corresponds to a Screen. Finally, behaviours that constrain the training environment correspond to standardized structural elements (e.g., a standard backboard and standard support frame), which impose boundary conditions on subsequent design.

After mapping functions to structures using the FBS framework, the corresponding key structural modules of the product are summarized in

Table 9. Structural Module.

Serial	Structure	Function	Example
S1	Camera Module	Monitor, record, detect, and acquire data.
S2	Positioning Marker	Enable basketball position localization within a small fixed environment.
S3	Ball Collection Net	Collect and retrieve missed basketballs.
S4	Processing Unit	Process historical data and generate training recommendations.
S5	Scoreboard Module	Record scores in real time.
S6	Display Screen	Display scoring information.
S7	Start Button	Enable emergency shutdown and rapid system startup.
S8	QR Code Scanner	Enable rapid user login and device status verification.
S9	Ball Launching Mechanism	Propel basketballs via roller wheels and regulate launch velocity by adjusting wheel rotational speed.
S10	Ball Storage Unit	Store basketballs and interface with the launching mechanism to enable a continuous collection–launch cycle.
S11	Rotating Base Platform	Rotate via a two-degree-of-freedom servo system and integrate with the launching mechanism to adjust the ball projection angle.
S12	Launch Direction Indicator Light	Indicate launch direction through controlled flashing sequences.
S13	Power Supply	Provide electrical power to the system.
S14	Speaker Module	Enhance system interactivity and operational usability.
S15	Guidance Light	Assist player movement through visual light guidance.
S16	Standard Backboard	Simulate real-game training conditions and facilitate procurement and assembly.
S17	Standard Support Frame	Simulate real-game training conditions and facilitate procurement and assembly.
S18	Protective Net	Block ball impact and protect the equipment.
S19	Main load-bearing frame	To serve as the primary load-bearing structure of the system, resisting ball impacts and protecting internal components.
S20	Ball guide chute	To guide the basketball into the Ball Storage Unit.
S21	Base	To support the overall weight of the machine.
S22	Telescopic support rod	To enable height adjustment and position locking of key components.

.

5. Design Proposal Presentation

5.1. Structural Configuration and Optimization

Based on functional characteristics, the structural system of the basketball shooting training machine was divided into five categories: execution, sensing, control, interaction, and support modules. Their corresponding structural codes are as follows: execution modules (S₃, S₉, S₁₁, S₂₀), sensing modules (S₁, S₂), control modules (S₄), interaction modules (S₅, S₈, S₁₂, S₁₄, S₁₅), and support modules (S₁₃, S₁₆, S₁₇, S₁₈, S₁₉, S₂₁, S₂₂), as illustrated in Figure 10.

To satisfy the requirements of basketball training scenarios, a rotating chassis was installed beneath the ball-launching unit. Directional indicator lights were positioned on the upper and middle sections of the device to guide players’ movement during training. Protective panels and a ball-collection net were mounted around the machine to shield the rotating chassis and launching unit from direct ball impact. Basketballs are guided by the upper collection net into the ball track and then transferred to the storage unit. The launching mechanism subsequently propels the ball through roller friction. By adjusting the roller speed and the rotation angle of the chassis, the machine can generate different launch speeds and angles to meet diverse training needs.

5.2. Presentation of Design Alternatives

Following the function-to-structure mapping established via the Function–Behavior–Structure (FBS) framework, design software was employed to formalize design concepts. Three candidate design alternatives (Alternatives A, B, and C) were developed in strict alignment with the empirically derived hierarchical user requirement framework, as illustrated in Figure 11.

Alternative A is characterized by a flat-panel structural form and integrated full-width light bar assemblies. Its streamlined monocoque structure and clear disassembly and assembly pathways for all components enable reduced manufacturing and maintenance costs. The full-width light bars serve a dual functional purpose, providing both ball launch direction indication and real-time system status feedback, making this alternative suitable for application scenarios with stringent requirements for cost control and operational reliability.

Alternative B features a geometric form defined by a combination of planar and beveled surfaces. Protective mesh panels replace the side aluminum alloy plates to reduce overall system weight and production costs, while the side mesh design reinforces the structural rigidity of the main frame. Narrow-profile light strips are integrated along the frame perimeter to deliver ball launch direction guidance, endowing the overall form with high structural integrity and a distinct visual identity.

Alternative C is designed around the core design principle of training function centralization. The ball launch direction indicator light array is centrally mounted at the top of the system, with all core training and performance feedback functions integrated into the upper section of the device, and primary decorative form elements positioned in the lower section. This layout is engineered to direct the user’s visual attention toward the core training task, minimizing distraction during practice.

5.3. Optimal Selection of Design Schemes

To enable the systematic optimization and selection of design alternatives for the basketball shooting training machine, latent variables that exert statistically significant or marginally significant effects on users’ continuance intention in the structural equation model (SEM) were selected as evaluation criteria, namely Performance Expectancy (PE), Effort Expectancy (EE), Task–Technology Fit (TTF), Facilitating Conditions (FC), and Social Influence (SI). Rating data for Alternatives A, B, and C against each evaluation criterion were collected via questionnaire surveys and expert scoring. The arithmetic mean of ratings for the same criterion under the same alternative was calculated to construct the original evaluation matrix for design alternatives, as presented in

Table 10. Evaluation of the original matrix and indicator weights in the design scheme.

Criterion	Path	Direct Effect	Indirect Effect	Total Effect	Normalized Weight	A	B	C
Performance Expectancy (PE)	0.440	0.440	0	0.440	0.260	8.4	6.3	7.9
Effort Expectancy (EE)	0.240	0.240	0	0.240	0.142	5.9	7.2	8.6
Task–Technology Fit (TTF)	0.320	0.320	0.405	0.725	0.428	8.5	6.1	7.5
Facilitating Conditions (FC)	0.100	0.100	0	0.100	0.059	7.0	8.3	5.8
Social Influence (SI)	0.190	0.190	0	0.190	0.112	7.3	7.5	6.8

.

On this basis, the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) was introduced to rank the design alternatives. All evaluation criteria were defined as benefit-type indicators, where higher values correspond to superior comprehensive performance of the alternative. The positive and negative ideal solutions were determined according to the weighted normalized evaluation results:

The positive ideal solution (Q⁺) and negative ideal solution (Q⁻) are calculated as follows:

$$Q^{+}=\left\{q_1^{+},q_2^{+},\cdots ,q_n^{+}\right\}=\left\{\left(\max q_{ij}\mid j\in R^{+}\right),\left(\min q_{ij}\mid j\in R^{-}\right)\right\}$$

(1)

Calculate the Euclidean distances to the positive and negative ideal solutions:

$$S_i^{+}=\sqrt{{\displaystyle \sum }_{j=1}^n{\left(q_{ij}-q_j^{+}\right)}^2} S_i^{-}=\sqrt{{\displaystyle \sum }_{j=1}^n{\left(q_{ij}-q_j^{-}\right)}^2}$$

(2)

Determine the relative closeness coefficient:

$$Z_i={\displaystyle \frac{S_i^{-}}{S_i^{+}+S_i^{-}}}\left(0\le Z_i\le 1\right)$$

(3)

The design alternatives were ranked according to their relative closeness coefficient (Zi), where a larger Zi value indicates a closer alignment of the alternative with the positive ideal solution in the TOPSIS framework. To ensure the objectivity and scientific rigor of the decision-making process and eliminate subjective bias in indicator weighting, the core determinants of user continuance intention identified via the Structural Equation Model (SEM) were adopted as the evaluation criteria for the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) analysis.

A total of 47 respondents were recruited to evaluate the three design alternatives, including 8 basketball coaches, 23 basketball enthusiasts, and 16 university students without prior basketball training experience. Respondents rated their perceived acceptance of each alternative against the five evaluation criteria using a continuous 0–10 rating scale, with scale anchors defined as follows: 0–3 = strong non-acceptance; 3–5 = relatively low acceptance; 5–6 = neutral acceptance; 6–8 = relatively high acceptance; 8–10 = strong acceptance.

Task–Technology Fit (TTF) not only had a direct effect on continuance intention (β = 0.32), but also exerted indirect effects through Performance Expectancy (TTF → PE = 0.61; PE → CI = 0.44) and Effort Expectancy (TTF → EE = 0.57; EE → CI = 0.24). As a result, its total effect reached 0.725. After normalization, the weights of the five evaluation indicators were as follows: TTF = 0.428, PE = 0.260, EE = 0.142, SI = 0.112, and FC = 0.059, as shown in Table 10.

Based on the mean ratings of 47 respondents for the three design alternatives across the five indicators, the TOPSIS analysis was recalculated. The results showed that Scheme A had the highest relative closeness coefficient (C_i = 0.745), followed by Scheme C (C_i = 0.632) and Scheme B (C_i = 0.179), as presented in

Table 11. Design scheme TOPSIS ranking result.

Scheme	S+	S−	Ci	Ranking
A	0.031	0.090	0.745	1
B	0.091	0.020	0.179	3
C	0.037	0.064	0.632	2

. After accounting for both the direct and indirect effects of TTF, Scheme A was closest to the positive ideal solution and was therefore identified as the optimal conceptual design under the current evaluation conditions. However, given the relatively limited evaluation sample and the heterogeneity of the respondents’ backgrounds, this result still requires further validation through subsequent prototype testing and subgroup comparison studies.

6. Discussion

6.1. Theoretical Contributions

This study makes two pivotal theoretical contributions to the existing literature:

First, it extends the applicability of the integrated UTAUT-TTF framework to the under-researched domain of automated sports training equipment. While the UTAUT model has been extensively validated in information systems and digital health research, its explanatory power in the context of self-directed sports training equipment remains largely unexamined. By integrating TTF theory and extending the model to include the TTF → PE and TTF → EE pathways, this study establishes a more comprehensive theoretical framework for explaining users’ sustained engagement with automated training devices, addressing a critical theoretical gap in sports equipment research between technical performance optimization and user behavior analysis.

Second, it establishes a systematic, evidence-based design decision-making paradigm that bridges empirical user behavior research and engineering design practice. To date, the potential of behavioral models to serve as engineering-level decision support tools has remained largely untapped. We translate statistically significant path coefficients into quantifiable user requirement weights, and we further operationalize these priorities via Function Analysis System Technique (FAST) functional decomposition and Function–Behavior–Structure (FBS) mapping. This provides a fully traceable theoretical pathway from granular user behavioral insights to actionable engineering design decisions.

6.2. Practical and Industrial Implications

The findings of this study deliver actionable, evidence-based guidance for the design, development, and optimization of automated sports training equipment:

First, product developers should prioritize functional configurations that directly enhance users’ performance expectancy. Core design focus should be placed on data-driven shooting performance feedback, game-like trajectory simulation that replicates real match scenarios, and personalized training plan generation, to ensure users perceive tangible improvements in athletic proficiency even over short-term usage.

Second, the usability and operational simplicity of the device must be treated as a non-negotiable core design priority. The robust, significant effect of effort expectancy underscores the critical need for intuitive, streamlined interaction logic and real-time interactive guidance to minimize the learning and operational burden for amateur users without professional equipment operation experience.

Third, task–technology fit must be fully integrated into the earliest stages of the design process. Functional design must be closely aligned with the core training tasks of basketball enthusiasts, including multi-angle shooting practice, dynamic movement training, and shooting form deviation correction, to ensure the technical capabilities of the device genuinely match users’ practical training needs.

Finally, social features should be positioned as supplementary optimization functions rather than core selling points. While social influence and facilitating conditions exhibit significant positive effects, their relative importance is markedly lower than that of performance and effort expectancy. Accordingly, lightweight social features—such as training result sharing and peer performance leaderboards—are sufficient, eliminating the need for overly complex social ecosystem construction.

6.3. Limitations and Future Research Directions

This study has several limitations. First, the sample primarily comes from a few urban areas in China and consists mainly of amateur basketball enthusiasts. Therefore, caution is needed when generalizing the findings to other regions and user groups. Second, the study is based on cross-sectional data, which lacks long-term tracking of changes in users’ continuance intention. Furthermore, the proposed design is still at the concept selection stage and has not yet been validated through prototype testing or real-world experiments to assess its usability and training effectiveness. Finally, the sample size for the TOPSIS evaluation is relatively small, and there is diversity in participant backgrounds, which may affect the stability of the results. Future research should expand the sample size, incorporate group comparisons, and adopt longitudinal designs, along with prototype testing, to enhance the robustness and practical value of the findings.

7. Conclusions

Based on the integrated UTAUT and TTF models, and combining prototype experience interviews with 429 valid questionnaires, this study systematically analyzes the key factors influencing users’ continuance intention to use the basketball shooting training machine and further translates the behavioral mechanisms into design decision paths. The results indicate that Performance Expectancy, Task–Technology Fit, and Effort Expectancy are the core factors influencing continuance intention. Among these, Performance Expectancy has the strongest impact, suggesting that users’ perception of clear improvement in training outcomes is a key determinant of their long-term usage intention.

Further analysis reveals that Task–Technology Fit not only directly impacts continuance intention but also exerts an indirect effect by enhancing Performance Expectancy and Effort Expectancy. This indicates that Task–Technology Fit plays a bridging role between training tasks, user cognition, and continuance behavior. In addition, both Technology Characteristics and Task Characteristics significantly influence Task–Technology Fit, indicating that the functional configuration of the training equipment should be centered around real training tasks.

On the design translation front, this study converts behavioral variables into engineering design criteria, forming a design solution that has been compared and evaluated. The results show that Solution A performs best in the comprehensive evaluation, with its advantages mainly stemming from a function-focused and task-oriented layout. This aligns with the research findings that Performance Expectancy and Task Fit are the core driving factors. It should be noted that although the overall model fit is acceptable, some indices still leave room for improvement. Therefore, this study is better suited as an explanatory framework rather than a mature predictive model.