ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 290

www.rsisinternational.org

Predicting Smart Supply Chain Performance Using Big Data

Analytics: A PLS-SEM and Machine Learning Hybrid Approach

Nor Ratna Masrom

, Wan Hasrulnizzam Wan Mahmood

, Al Amin Mohamed Sultan

1,2,3

Fakulti Pengurusan Teknologi dan Teknousahawanan, Universiti Teknikal Malaysia Melaka, 76100,

Melaka, Malaysia

*Corresponding Author

DOI: https://dx.doi.org/10.47772/IJRISS.2025.92800028

Received: 22 November 2025; Accepted: 28 November 2025; Published: 19 December 2025

ABSTRACT

This study proposes a hybrid approach integrating Partial Least Squares Structural Equation Modeling (PLS-

SEM) and Machine Learning (ML) techniques to predict Smart Supply Chain Management (SmSCM)

performance based on Big Data Analytics (BDA) adoption. While previous studies validated behavioral models,

this research advances predictive capabilities by leveraging both structural path analysis and data-driven

classification. The conceptual model is grounded in the UTAUT2 framework, incorporating constructs such as

Performance Expectancy, Effort Expectancy, Facilitating Conditions, Price Value, Perceived Risk, Technology

Readiness, and Trust. Data collected from 309 Malaysian manufacturing firms were first analysed using PLS-

SEM to confirm causal relationships and model reliability. Subsequently, supervised learning models which are

Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbours (KNN). These models were

applied to predict SmSCM performance classes (High vs Low) using behavioral and readiness indicators as

input features. Results indicate that combining PLS-SEM with ML enhances explanatory and predictive power,

with SVM outperforming other classifiers at 70.66% accuracy using entropy-informed features. This study

demonstrates the potential of hybrid analytics to guide data-driven decision-making in Industry 4.0 supply

chains. It contributes both theoretically and practically by offering a validated, predictive framework for BDA-

driven supply chain transformation

Keywords: Big Data Analytics; Smart Supply Chain; Technology Adoption; Machine Learning; Predictive

Modelling; Manufacturing Firm.

INTRODUCTION

The increasing complexity, interconnectedness, and volatility of global supply chains have amplified the need

for intelligent, data-driven systems capable of enhancing visibility, resilience, and responsiveness. In this

context, manufacturing firms are increasingly adopting Big Data Analytics (BDA) as a strategic enabler for

operational optimization. BDA provides predictive and prescriptive capabilities that allow firms to optimize

inventory management, improve demand forecasting, anticipate disruptions, and enhance delivery accuracy

(Wamba et al., 2020). By leveraging large-scale, real-time datasets, BDA enables supply chains to shift from

reactive problem-solving toward proactive decision-making.

While the theoretical benefits of BDA are well-documented, the extent to which adoption translates into

measurable improvements in Smart Supply Chain Management (SmSCM) performance remains insufficiently

explored—particularly in emerging economies such as Malaysia. SmSCM involves the integration of advanced

digital technologies, such as the Internet of Things (IoT), Artificial Intelligence (AI), and BDA, to enhance

supply chain performance dimensions including visibility, agility, collaboration, and decision quality. Although

BDA adoption has been linked to performance improvement in studies conducted in digitally mature economies,

evidence from resource-constrained environments remains scarce. This gap is critical, as the enabling conditions,

risk perceptions, and readiness levels in emerging markets differ significantly from those in advanced economies.

Existing research on technology adoption in supply chain contexts has often drawn upon models such as the

Unified Theory of Acceptance and Use of Technology 2 (UTAUT2) to identify behavioral, organizational, and

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 291

www.rsisinternational.org

technological determinants of adoption. While these models offer robust explanatory power, they are primarily

designed to test causal relationships, focusing on constructs such as Performance

Expectancy (PE), Effort Expectancy (EE), Facilitating Conditions (FC), and Price Value (PV). However, these

models rarely assess predictive performance—that is, the ability to forecast which firms are most likely to

achieve high SmSCM outcomes based on adoption behavior. This lack of predictive integration limits the

practical applicability of behavioral frameworks for strategic decision-making in industry.

Furthermore, prior studies often overlook con-textual variables that influence technology adoption in emerging

economies, such as Perceived Risk (PR), Trust, and Technology Readiness (TR). Perceived Risk is particularly

relevant in environments where concerns about data security, vendor reliability, and return on investment are

heightened. Trust, both in technology providers and in system integrity, plays a pivotal role in mitigating

adoption hesitation, especially when supply chain operations involve sensitive transactional and operational data.

Technology Readiness captures the extent to which individuals and organizations possess the skills, mindset,

and infrastructure to embrace technological change—an enabler often absents in models focusing solely on

behavioral intention.

To address these gaps, the present study pro-poses a hybrid methodological approach that integrates Partial Least

Squares Structural Equation Modelling (PLS-SEM) with super-vised machine learning (ML) techniques to both

explain and predict SmSCM performance. The extended UTAUT2 model adopted in this research incorporates

PR, Trust, and TR as domain-specific constructs, while excluding Social Influence and Hedonic Motivation,

which have limited relevance in B2B manufacturing contexts where adoption is typically mandated by top

management rather than driven by peer influence or user enjoyment. This tailoring enhances the contextual

validity of the model for Malaysian manufacturing firms.

From a methodological perspective, PLS-SEM is employed to validate the extended UTAUT2 framework and

assess the structural relationships between adoption determinants, Usage Behavior (UB), and SmSCM

performance. Subsequently, the validated constructs are used as input features in ML classification models—

Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and Random Forest (RF)—to predict whether

firms are likely to achieve high or low SmSCM performance. This dual-stage process not only tests theoretical

relationships but also applies predictive analytics to produce actionable classification outcomes. The ML

component employs Leave-One-Out Cross-Validation (LOOCV) to address the moderate dataset size, ensuring

robust model evaluation while mitigating overfitting risks.

The novelty of this research is threefold:

First integration of UTAUT2, PLS-SEM, and ML for SmSCM performance prediction in an emerging market

context. While previous studies have separately applied adoption models or predictive analytics, none have

combined these approaches to bridge the gap between behavioral explanation and performance prediction.

Context-specific extension of UTAUT2 by incorporating PR, Trust, and TR, addressing adoption barriers

particularly relevant to manufacturing firms in resource-constrained environments. This adaptation responds to

calls by Dwivedi et al. (2022) for greater contextualization of adoption frameworks.

Direct linkage between usage behavior and SmSCM performance, operationalized through multi-dimensional

performance metrics, there-by addressing the gap identified by Zhou et al. (2020) that most adoption studies stop

short of measuring downstream operational outcomes.

By combining theoretical rigor with predictive modelling, this study makes a dual contribution. Theoretically, it

enhances the explanatory power of UTAUT2 for BDA adoption in emerging economies, demonstrating that

Trust and TR may outweigh traditional cost–benefit considerations (Price Value) in driving adoption behavior.

Practically, it equips managers with predictive tools capable of identifying performance improvement

opportunities. For example, if the SVM classifier predicts low performance for a given firm, targeted

interventions—such as TR-enhancement training, infrastructure upgrades, or trust-building measures—can be

implemented to address the underlying behavioral and organizational gaps.

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 292

www.rsisinternational.org

This hybrid approach is especially relevant under the Industry 4.0 transformation agenda, where data-driven

decision-making is both a strategic necessity and a competitive differentiator. In emerging markets, where

resource allocation must be carefully prioritized, the ability to predict which adoption pathways will yield the

highest performance returns offers significant managerial value. Moreover, policymakers can leverage such

insights to design interventions—such as government subsidies for BDA infrastructure or national analytics

training programmed—that address systemic barriers and accelerate adoption across the manufacturing sector.

In sum, this research responds to the call for integrated, context-aware adoption models that go beyond

explanation to deliver predictive insights. By applying an extended UTAUT2 model within a PLS-SEM and ML

hybrid framework, it offers a novel, empirically vali-dated approach to understanding and forecasting SmSCM

performance in Malaysian manufacturing. The outcomes of this study are intended to inform both scholarly

discourse and practical strategy, providing a replicable template for similar investigations in other emerging

market contexts.

LITERATURE REVIEW

Big Data Analytics (BDA) has emerged as a transformative enabler for converting traditional supply chains into

intelligent and responsive networks. Akter et al. (2016) argued that BDA capabilities, when aligned with business

strategy, enhance firm performance through improved decision-making and process efficiency. In

manufacturing, BDA supports real-time monitoring, predictive forecasting, and enhanced collaboration,

contributing to agility and resilience (Wamba et al., 2020). In developing countries, Aghimien et al. (2021) noted

that organizational culture, top management support, and technological infrastructure significantly influence

readiness for BDA adoption.

Dubey et al. (2019) further highlighted that competitive advantage is achievable when analytical insights are

successfully translated into operational improvements. However, adoption among small and medium-sized

enterprises (SMEs) remains constrained by cost, skill shortages, and inadequate infrastructure (Mandal, 2017).

This indicates that BDA adoption must be assessed not only from a technological standpoint but also in terms of

organizational readiness and strategic intent—particularly in emerging economies like Malaysia.

A. Theoretical Gaps in BDA Adoption Studies

Despite the growing literature on BDA adoption, a persistent gap exists between behavioral intention and

measurable performance out-comes. For example, Zhou et al. (2020) examined behavioral predictors of BDA

adoption but did not connect them to tangible supply chain performance metrics, limiting their practical value.

Similarly, Dwivedi et al. (2022) extended the UTAUT2 model to include Trust and Risk but did not explore the

predictive capabilities of these constructs through advanced analytics. This study addresses these limitations by

linking usage behavior directly to Smart Supply Chain Management (SmSCM) performance and employing a

hybrid methodology that combines PLS-SEM for causal ex-planation with machine learning (ML) for predictive

validation.

B. Justification for Using UTAUT2 and Dropping Certain Constructs

The Unified Theory of Acceptance and Use of Technology 2 (UTAUT2) was selected as the core framework for

this study due to its ability to integrate behavioral, organizational, and contextual determinants of technology

usage (Venkatesh et al., 2012). Unlike the Technology–Organization–Environment (TOE) framework, which is

more suited to analyzing structural readiness (Baker, 2012), UTAUT2 captures individual-level adoption drivers

critical to BDA contexts. The Diffusion of Innovation (DOI) theory offers innovation attributes like relative

advantage and compatibility but falls short in explaining post-adoption behaviors and trust-based dynamics

essential to sustained BDA use.

However, UTAUT2 in its original form contains constructs—Social Influence and Hedonic Motivation—that

are less applicable to B2B industrial settings. Social Influence, which captures peer pressure or societal

expectations, has limited relevance where adoption is mandated by top management rather than peer consensus

(Alalwan et al., 2020). Hedonic Motivation, centered on enjoyment, is misaligned with the utilitarian and

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 293

www.rsisinternational.org

performance-driven nature of BDA systems. Removing these constructs im-proves contextual fit and ensures

that the model focuses on variables directly relevant to performance-driven adoption.

To strengthen explanatory power, this study incorporates Technology Readiness (TR), Perceived Risk (PR), and

Trust. TR, first conceptualized by Parasuraman (2000), bridges user disposition with actual usage and has shown

predictive value in BDA contexts (Lin et al., 2020). PR addresses uncertainty in outcomes, while Trust has been

identified as a critical enabler in data-intensive environments (Kshetri, 2021). Together, these additions adapt

UTAUT2 to the realities of BDA adoption in Malaysian manufacturing.

C. Malaysian Manufacturing Context

The Malaysian manufacturing sector presents unique adoption challenges. According to the Khazanah Research

Institute (2021), SMEs face persistent barriers, including limited digital infrastructure, a shortage of skilled

talent, and constrained budgets. While policies like MITI’s Industry4WRD aim to accelerate transformation,

uptake varies widely across regions and firm sizes. Mandal (2017) found that many firms lack the internal

capabilities to implement BDA without external assistance, leading to slower adoption.

Lee et al. (2019) and Aghimien et al. (2021) observed that while Industry 4.0 awareness is improving, strategic

alignment remains weak, resulting in fragmented implementation. These conditions necessitate an adoption

model that incorporates readiness, risk, and trust—factors often overlooked in generic models but critical for

BDA adoption in Malaysia.

D. BDA in Emerging vs Advanced Economies

Contrasts between advanced and emerging economies highlight the importance of context-specific modelling.

Wamba et al. (2020) reported significant performance gains from BDA adoption in European manufacturers,

supported by robust infrastructure and mature analytics capabilities. In contrast, firms in emerging economies

often face institutional voids, weak data governance, and interoperability issues.

For example, Akter et al. (2016) noted that in Bangladesh, successful BDA adoption was concentrated in

multinational subsidiaries with external support, limiting scalability. This suggests that models for emerging

economies must account for environmental constraints. The inclusion of PR and TR in the current study

addresses these gaps, offering nuanced insights into adoption drivers under less favorable conditions.

E. Predictive Analytics, Machine Learning, and the Hybrid Approach

Machine learning has gained traction in supply chain research for its ability to identify nonlinear patterns and

improve forecasting accuracy. Algorithms like SVM, RF, and KNN have been applied to tasks such as delay

prediction, sup-plier risk assessment, and performance classification (Chong et al., 2017; Wang et al., 2022).

Wichmann et al. (2021) demonstrated the effectiveness of SVM in classifying disruptions using IoT data,

underscoring the potential for ML in real-time supply chain decision-making.

However, while ML excels in prediction, it often lacks theoretical grounding. Hybrid approaches that combine

PLS-SEM’s causal explanation with ML’s predictive capability are gaining momentum. Wang et al. (2022)

showed that integrating SEM with Random Forest im-proved delivery delay predictions, while Chong et al.

(2017) found that hybrid models enhanced fraud detection accuracy in supply chain finance.

In the present study, PLS-SEM confirmed Usage Behavior as a significant mediator (β = 0.341) between

antecedents and SmSCM performance, but SVM achieved only moderate predictive accuracy (70.66%). This

divergence reinforces that statistical significance does not always translate to predictive sufficiency, making the

hybrid approach both theoretically and practically valuable.

By embedding predictive analytics into a validated behavioral framework, this study provides a dual-function

model—explaining adoption behavior while forecasting performance outcomes—offering both academic

contribution and managerial utility.

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 294

www.rsisinternational.org

METHODOLOGY

By adopting a hybrid methodology, this study offers a comprehensive lens: PLS-SEM captures the structural

relationships among UTAUT2 constructs, while ML algorithms evaluate the extent to which these constructs

can accurately classify firms into high or low SmSCM performance categories. This approach not only enriches

the academic discourse but also delivers practical value to decision-makers seeking both diagnostic and

prescriptive tools for supply chain transformation.

Furthermore, this integration aligns with the broader shift in supply chain research toward analytics-driven

models. As Ivanov and Dolgui (2020) point out, the post-COVID-19 era demands new frameworks that can

predict disruptions, adapt quickly, and sustain performance. In this context, hybrid models such as the one

proposed in this study offer the dual advantage of understanding "why" a phenomenon occurs and "what" will

likely happen next.

This study adopts a hybrid methodological approach integrating Partial Least Squares Structural Equation

Modelling (PLS-SEM) and supervised machine learning (ML) to predict Smart Supply Chain Management

(SmSCM) performance from Big Data Analytics (BDA) adoption constructs. The methodology comprises three

main phases: instrument validation, structural model estimation, and predictive modelling.

A. Instrument Development

The initial phase involved developing a survey instrument based on the UTAUT2 framework, incorporating

Performance Expectancy (PE), Effort Expectancy (EE), Facilitating Conditions (FC), Price Value (PV),

Perceived Risk (PR), Technology Readiness (TR), Trust, BDA Usage Behavior (UB), and SmSCM

Performance. Content validity was established through expert panel review using the Content Validity Index

(CVI). Construct reliability and validity were assessed through PLS-SEM using SmartPLS 4.0, evaluating

indicator reliability, internal consistency (Cronbach’s alpha, Composite Reliability), convergent validity

(Average Variance Extracted), and discriminant validity (Fornell-Larcker and HTMT criteria).

B. Data Collection and Sample

Data were collected via an online questionnaire distributed to Malaysian manufacturing firms listed in industry

directories and supply chain networks. A total of 309 usable responses were obtained, satisfying the

recommended mini-mum sample size for PLS-SEM. Respondents included mid- to senior-level managers

involved in digital transformation or supply chain operations. The dataset was screened for missing values and

outliers before analysis.

C. Structural Equation Modelling

PLS-SEM was employed to estimate the causal relationships among the constructs. Bootstrapping with 5,000

resamples was conducted to assess path coefficients, significance levels, and predictive relevance (Q²). The

model’s explanatory power was evaluated using R² values for endogenous constructs. This step confirmed the

underlying behavioral structure of BDA adoption and its influence on SmSCM performance.

D. Machine Learning Classification

To complement the structural model and assess predictive accuracy, a supervised machine learning (ML)

classification task was conducted using the validated constructs from the PLS-SEM analysis as input features.

The target variable, Smart Supply Chain Management (SmSCM) performance, was converted into binary classes

(High vs Low) using a median split approach, which provides a balanced distribution and supports

interpretability.

Three algorithms were evaluated: Support Vec-tor Machine (SVM), Random Forest

(RF), and K-Nearest Neighbours (KNN). Given the moderate sample size (n = 309),

Leave-One-Out Cross-Validation (LOOCV) was selected as the validation

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 295

www.rsisinternational.org

strategy. LOOCV ensures that each observation is used once as a test case while

the remaining n − 1 samples serve as the training set. This method maximizes

data utilization and produces a nearly unbiased estimate of generalization

error, making it particularly suitable for behavioral studies with limited data.

To ensure reproducibility and optimal performance, hyperparameter tuning was carried out for each algorithm.

For the SVM classifier, a Radial Basis Function (RBF) kernel was ap-plied, with a grid search used to identify

the optimal values of the regularization parameter (C) and kernel coefficient (γ). For Random Forest, the number

of estimators (n_estimators) and maximum tree depth (max_depth) were tuned to minimize overfitting and

enhance generalisation. KNN was included as a baseline due to its simplicity and interpretability, with k values

ranging from 3 to 9 tested.

Model performance was evaluated using standard classification metrics: accuracy, macro precision, macro recall,

and macro F1-score. These metrics were selected to reflect the model’s ability to classify both classes equally

well, particularly important in datasets with class imbalance or near-threshold groupings. Confusion matrices

were also analyzed to identify misclassification patterns and evaluate model reliability at the class level.

All experiments were implemented using Py-thon's scikit-learn library, with fixed random seeds to ensure

deterministic outputs. This ML component enhances the study’s contribution by extending beyond structural

path modelling, providing a predictive layer that supports data-driven decision-making in supply chain contexts.

E. Software and Tools

PLS-SEM analysis was conducted using SmartPLS 4.0. Machine learning models were implemented in Python

using scikit-learn. All preprocessing, model tuning, and evaluation were performed using standardized pipelines

to ensure reproducibility.

RESULT

This section presents the empirical results obtained from both the PLS-SEM analysis and the machine learning

classification phase. The integrated approach enables both explanatory validation of the proposed model and

predictive assessment of Smart Supply Chain Management (SmSCM) performance based on Big Data Analytics

(BDA) adoption indicators

A. Measurement Model Evaluation

The reliability and validity of the constructs were first examined. All items exhibited outer loadings above 0.70,

indicating satisfactory indicator reliability. Composite Reliability (CR) values exceeded the 0.70 threshold for

all constructs, demonstrating internal consistency. Average Variance Extracted (AVE) scores surpassed the 0.50

criterion, confirming convergent validity. Discriminant validity was verified using both Fornell–Larcker and

Heterotrait–Monotrait (HTMT) ratios, with all values below the acceptable threshold of 0.85. These results

affirm the robustness of the measurement mode

B. Structural Model Assessment

Bootstrapping with 5,000 resamples was used to evaluate the significance of path coefficients. Significant

relationships were found between Effort Expectancy (β = 0.224, p < 0.01), Facilitating Conditions (β = 0.207, p

< 0.01), and Trust (β = 0.256, p < 0.001) with BDA Usage Behavior. In turn, Usage Behavior significantly

predicted SmSCM Performance (β = 0.341, p < 0.001). The R² value for SmSCM Performance was 0.486,

indicating that nearly 49% of the variance is explained by the proposed model. The model exhibited strong

predictive relevance (Q² = 0.329), suggesting adequate out-of-sample predictive capability.

C. Machine Learning Classification Results

Table 1 presents the classification results using three supervised machine learning models—Support Vector

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 296

www.rsisinternational.org

Machine (SVM), Random Forest (RF), and K-Nearest Neighbours (KNN) to Therefore, beyond overall

accuracy, macro-averaged precision, recall, and F1-score were used to ensure equal weighting of both classes.

Future work may consider incorporating ROC-AUC to assess classifier discrimination capability in the presence

of subtle class imbalance. Additionally, stratified sampling or resampling techniques (e.g., SMOTE) could be

employed to enhance predict High vs Low SmSCM Performance, replacing earlier ambiguous class labels such

as Pre-Frail or Pre-CF. Among the models tested, SVM consistently outperformed RF and KNN, achieving the

highest accuracy (70.66%) and macro F1-score (68.20%). This superior performance can be attributed to SVM’s

robust-ness in handling smaller datasets and its ability to maximise the margin between classes, making it

effective in high-dimensional, low-sample-size contexts.

In contrast, RF demonstrated signs of overfitting, likely due to its ensemble nature and sensitivity to noise when

the number of features is comparable to the sample size. KNN, while computationally simple and interpretable,

suffered from limited generalization capacity in this scenario, particularly due to sensitivity to localized data

variation and potential noise amplification during neighbor selection.

TABLE I COMPARATIVE PERFORMANCE OF ML MODELS FOR PREDICTING SMSCM PERFORMANCE (HIGH VS

LOW)

Model

Accuracy (%)

Macro Precision (%)

Macro Recall

(%)

Macro F1-

Score (%)

Support Vector Machine

(SVM)

70.66

67.5

68.9

68.2

Random Forest (RF)

54.05

52.8

53.6

53.1

K-Nearest Neighbours (KNN)

56.76

55.4

56.2

55.8

Notes:

● SVM outperformed RF and KNN across all metrics.

● Macro-averaged metrics ensure fair comparison despite slight class imbalance.

● LOOCV was used for model validation due to moderate sample size (n = 309).

Class imbalance was also examined. Although the binary grouping based on median split produced relatively

balanced class sizes, minor skewness may still affect metric interpretation. Therefore, beyond overall accuracy,

macro-averaged precision, recall, and F1-score were used to ensure equal weighting of both classes. Future work

may consider incorporating ROC-AUC to assess classifier discrimination capability in the presence of subtle

class imbalance. Additionally, stratified sampling or resampling techniques (e.g., SMOTE) could be employed

to enhance model fairness and sensitivity.

These findings underscore the potential of SVM-based prediction as a valuable tool for supporting BDA-driven

SmSCM decisions, especially when working with moderate sample sizes and behavioral datasets.

SUMMARY OF FINDINGS

The hybrid approach validates the proposed behavioral model while also demonstrating the feasibility of

predicting SmSCM outcomes using machine learning classifiers. The results suggest that combining statistical

inference with predictive analytics offers a more comprehensive assessment of BDA adoption effectiveness.

DISCUSSION

The classification results offer meaningful insights into the model’s limitations and opportunities for

enhancement. The observed F1-score disparity between High (80.90%) and Low SmSCM firms (50.00%), as

shown in Table 1, suggests that current behavioral constructs—such as Trust and Effort Expectancy—may be

more effective at characterizing high-performing firms than underperformers. This asymmetry could stem from

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 297

www.rsisinternational.org

either data imbalance or a lack of discriminative behavioral signals among Low SmSCM firms. To address this,

future research should explore the integration of operational metrics, such as inventory turnover, cycle time, or

on-time delivery rates, which may provide more granular indicators of supply chain underperformance and

enhance classification sensitivity.

Moreover, while the PLS-SEM results validated Usage Behavior (UB) as a significant predictor of SmSCM

performance (β = 0.341), the SVM model's moderate predictive accuracy (70.66%) suggests that UB alone may

not be sufficient for precise outcome prediction. This reinforces the idea that explanatory significance does not

necessarily equate to predictive power. Accordingly, hybrid models could benefit from the inclusion of firm-

level covariates, such as IT infrastructure maturity, number of digital touchpoints, or workforce size, to capture

structural or resource-based factors that influence SmSCM success beyond user perceptions and behavioral

intent.

Together, these insights encourage a broader, multi-dimensional approach to modelling technology adoption

outcomes, combining behavioral, operational, and contextual data streams to improve both theoretical fidelity

and real-world applicability.

A. Theoretical Implications

The present study advances the understanding of Big Data Analytics (BDA) adoption and Smart Supply Chain

Management (SmSCM) performance in Malaysian manufacturing by extending the Unified Theory of

Acceptance and Use of Technology 2 (UTAUT2) with three contextual constructs—Technology Readiness

(TR), Perceived Risk (PR), and Trust. The PLS-SEM analysis revealed that Effort Expectancy (EE), Facilitating

Conditions (FC), and Trust significantly influenced Usage Behavior (UB), while Price Value (PV) and Perceived

Risk (PR) were non-significant. This differs from Dwivedi et al. (2022), who found PV to be a strong driver of

adoption in broader digital transformation contexts. The divergence may stem from the Malaysian manufacturing

sec-tor’s prioritization of operational reliability and data security over cost considerations, particularly in BDA

initiatives where upfront investment is often mandated.

The finding that Trust exerts a stronger influence on UB than PV challenges traditional UTAUT2 assumptions,

which typically position cost–benefit perceptions as central to adoption decisions. In a context where BDA

systems handle sensitive operational and customer data, trust in system reliability, vendor integrity, and data

governance frameworks becomes pivotal. This aligns with Kshetri (2021), who emphasized that in emerging

economies, perceived trustworthiness can outweigh economic incentives in technology adoption decisions.

Technology Readiness emerged as a meaningful mediator, bridging the gap between antecedents and UB. This

supports Parasuraman’s (2000) original assertion that readiness is not merely a user trait but a strategic lever that

organizations can enhance through training, leadership support, and infrastructure investment. The integration

of TR into the UTAUT2 framework addresses criticisms regarding the original model’s limited focus on the

“capability to adopt” dimension, especially in resource-constrained environments.

The results also substantiate the conceptual link between UB and SmSCM performance, confirming that

behavioral adoption translates into measurable operational outcomes such as improved visibility, agility,

collaboration, and decision quality. This outcome addresses the gap identified by Zhou et al. (2020), who noted

that most behavioral adoption studies stop short of linking usage to downstream performance metrics.

Based on these findings, a revised UTAUT2 model for BDA adoption retains key UTAUT2 constructs (EE, PE,

FC, PV) but prioritizes Trust and TR, while excluding Hedonic Motivation and Social Influence. It also explicitly

connects UB to SmSCM performance, creating a bridge between behavioral theory and performance-based

frameworks. This configuration offers stronger explanatory power for emerging market contexts and a

foundation for hybrid explanatory–predictive analytics.

B. Practical Recommendations

Firms seeking to maximize SmSCM performance through BDA adoption should prioritise initiatives that build

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 298

www.rsisinternational.org

Trust and enhance Technology Readiness. This involves:

Trust building measures – Establishing robust data governance policies, obtaining internationally recognised

security certifications (e.g., ISO/IEC 27001), and maintaining transparent vendor relationships can reduce

apprehension over data misuse and system failures.

Technology Readiness training – Investing in targeted skills development, such as analytics literacy programs

and cross-functional digital competency workshops, can significantly improve UB. Given the SVM’s ability to

predict high SmSCM performance with 80.9% F1-score for the High category, enhancing TR could directly

improve predictive accuracy and real-world outcomes.

Facilitating Condition enhancement – Aligning IT infrastructure with BDA requirements, ensuring system

compatibility, and providing continuous technical support can remove adoption barriers and sustain engagement.

The machine learning results demonstrate that behavioral constructs can be leveraged for predictive performance

classification. Firms can deploy similar models internally to identify business units or plants with high potential

for performance improvement. This enables targeted interventions, such as prioritizing resource allocation or

implementing specific capability-building programs in low-performing areas.

C. Policy Implications

At a policy level, the findings indicate that national strategies for Industry 4.0 should place stronger emphasis

on reducing Perceived Risk and improving Technology Readiness. Possible interventions include:

Government subsidies for BDA infrastructure – Financial incentives, tax rebates, or low-interest loans for digital

infrastructure investment can lower the perceived financial and operational risk of adoption.

National training frameworks – Public–private partnerships could deliver sector-specific analytics and data

governance training, particularly for SMEs, which are most constrained by skill shortages.

Standardization and certification – Implementing national standards for BDA interoperability and security can

increase trust in technology systems and vendor solutions.

By aligning industrial policy with the key behavioral determinants identified in this study, policymakers can

accelerate adoption and performance outcomes across the manufacturing sector.

CONCLUSION

This study presents a hybrid methodological framework that integrates Partial Least Squares Structural Equation

Modelling (PLS-SEM) with supervised machine learning (ML) to predict Smart Supply Chain Management

(SmSCM) performance based on Big Data Analytics (BDA) adoption indicators. Grounded in the UTAUT2

framework and extended with constructs such as Technology Readiness, Trust, and Perceived Risk, the research

provides both explanatory insights and predictive capabilities within the context of Malaysian manufacturing

firms.

The PLS-SEM results validate the influence of key behavioral and organizational factors particularly Effort

Expectancy, Facilitating Conditions, and Trust on BDA usage behavior. Furthermore, BDA usage significantly

predicts SmSCM performance, reinforcing the strategic importance of analytics adoption in digital supply

chains. These findings not only confirm prior theoretical propositions but also offer empirical evidence specific

to emerging market contexts.

The application of ML, especially Support Vector Machine (SVM), demonstrates the feasibility of forecasting

supply chain performance based on behavioral indicators. The classification model achieved a predictive

accuracy of 70.66%, indicating that firm-level adoption characteristics can serve as reliable inputs for data-

driven performance assessment. This dual approach bridges the methodological gap between behavioral

modelling and out-come prediction, offering a robust decision-support framework for practitioners.

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)

ISSN: 2454-6186 | DOI: 10.47772/IJRISS

Special Issue | Volume IX Issue XXVIII November 2025

Page 299

www.rsisinternational.org

In sum, this study contributes to academic literature by proposing a validated, predictive model for BDA-driven

SmSCM transformation. For practitioners, the model provides a structured diagnostic and forecasting tool that

can support strategic investments in digital technologies. Future research may further refine this framework by

incorporating sector-specific variables, real-time analytics data, or physio-logical measures to enhance predictive

precision and contextual relevance.

While the study contributes to both theory and practice, certain limitations must be acknowledged. First, the

sample comprised primarily mid- to senior-level managers, potentially introducing positional bias toward

strategic perspectives, while operational insights from technical staff were underrepresented. Future research

should adopt a multilevel sampling approach to capture both strategic and operational viewpoints.

Second, the cross-sectional design limits the ability to infer causal relationships over time. A longitudinal study

tracking firms from adoption through to maturity could reveal how behavioral determinants evolve and interact

with performance metrics.

Third, while the binary classification of SmSCM performance via median split ensured balanced classes for ML

modelling, it may have oversimplified performance variation. Future work could adopt multi-class or regression-

based predictive modelling to capture a more nuanced performance spectrum.

Fourth, predictive modelling relied on behavioral constructs alone. Incorporating IoT-derived operational data—

such as real-time inventory turnover, lead time variability, and production yield—could improve prediction

accuracy, especially for Low-performance firms, which the current model struggled to classify (F1 = 50.0%).

Finally, while SVM emerged as the best-performing model in this study, further exploration of ensemble learning

techniques and explainable AI (XAI) approaches could yield both higher accuracy and greater interpretability,

making predictive analytics more actionable for decision-makers.

ACKNOWLEDGMENT

The authors would like to acknowledge the support provided by Universiti Teknikal Malaysia Melaka (UTeM)

for the facilities and academic resources that enabled the completion of this research. Special thanks are also

extend-ed to the participating Malaysian manufacturing firms for their time and insights during the data

collection

REFERENCES

1. Aronoff, S., (1989). Geographic Information Systems: A Management Perspective. Ottawa: WDL

Publications.

2. Dwivedi, Y. K., Rana, N. P., Jeyaraj, A., Clement, M., & Williams, M. D. (2022). Re-examining the

Unified Theory of Acceptance and Use of Technology (UTAUT): Towards a revised theoretical

model. Information Systems Frontiers, 24(1), 1–16. https://doi.org/10.1007/s10796-021-10141-5

3. Alalwan, A. A., Dwivedi, Y. K., & Rana, N. P. (2020). Empirical examination of UTAUT2 in the

context of mobile payment adoption. Information Systems Frontiers, 22(1), 293–

314. https://doi.org/10.1007/s10796-018-9862-7

4. Alshamaila, Y., Papagiannidis, S., & Li, F. (2013). Cloud computing adoption by SMEs in the UK: A

multi-perspective framework. Journal of Enterprise Information Management, 26(3), 250–

275. https://doi.org/10.1108/17410391311325225

5. Choi, T.-M., Wallace, S. W., & Wang, Y. (2018). Big data analytics in operations

management. Production and Operations Management, 27(10), 1868–

1883. https://doi.org/10.1111/poms.12838

6. Chong, A. Y.-L., Ch’ng, E., Liu, M. J., & Li, B. (2017). Predicting consumer product demands via

Big Data: The roles of online promotional marketing and online reviews. International Journal of

Production Research, 55(17), 5142–5156. https://doi.org/10.1080/00207543.2017.1292063

7. Frederico, G. F., Garza-Reyes, J. A., Anosike, A., & Kumar, V. (2021). Supply Chain 4.0: Concepts,

maturity, and research agenda. Supply Chain Management: An International Journal, 26(2), 262–

ICTMT 2025 | International Journal of Research and Innovation in Social Science (IJRISS)  
ISSN: 2454-6186 | DOI: 10.47772/IJRISS  
Special Issue | Volume IX Issue XXVIII November 2025 
    
 
 
 Page 300  
   www.rsisinternational.org 
282. https://doi.org/10.1108/SCM-09-2020-0454 
8.  Gupta, S., Modgil, S., & Gunasekaran, A. (2020). Big data in lean six sigma: A review and future 
research  directions. International  Journal  of  Production  Research,  58(3),  947–
969. https://doi.org/10.1080/00207543.2019.1598599 
9.  Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2022). A primer on partial least squares 
structural equation modeling (PLS-SEM) (3rd ed.). SAGE Publications. 
10. Hazen, B. T., Skipper, J. B., Ezell, J. D., & Boone, C. A. (2016). Big data and predictive analytics for 
supply chain sustainability: A theory-driven research agenda. Computers & Industrial Engineering, 
101, 592–598. https://doi.org/10.1016/j.cie.2016.06.030 
11. Ivanov, D., Dolgui, A., & Sokolov, B. (2022). Cloud supply chain: Integrating Industry 4.0 and digital 
platforms  in  the  “Supply  Chain-as-a-Service”. Transportation  Research  Part  E:  Logistics  and 
Transportation Review, 160, 102676. https://doi.org/10.1016/j.tre.2022.102676 
12. Kshetri, N. (2021). Blockchain and trust in supply chain management. In Blockchain and supply chain 
management (pp. 3–23). Springer. https://doi.org/10.1007/978-3-030-73659-4_1 
13. Kshetri, N. (2021). Blockchain and trust in supply chain management. In Blockchain and supply chain 
management (pp. 3–23). Springer. https://doi.org/10.1007/978-3-030-73659-4_1 
14. Liao, Y., Deschamps, F., Loures, E. D. F. R., & Ramos, L. F. P. (2017). Past, present and future of 
Industry 4.0: A systematic literature review and research agenda. International Journal of Production 
Research, 55(12), 3609–3629. https://doi.org/10.1080/00207543.2017.1308576 
15. Mittal, S., Khan, M. A., Romero, D., & Wuest, T. (2018). A critical review of smart manufacturing & 
Industry 4.0 maturity models: Implications for SMEs. Journal of Manufacturing Systems, 49, 194–
214. https://doi.org/10.1016/j.jmsy.2018.10.005 
16. Ngai,  E.  W.  T.,  &  Gunasekaran,  A.  (2020). Big  data  analytics  in  logistics  and  supply  chain 
management: Certain investigations for research and applications. International Journal of Production 
Economics, 176, 98–110. https://doi.org/10.1016/j.ijpe.2016.03.014 
17. Ooi,  K.  B.,  Lee,  V.  H.,  Tan,  G.  W.  H.,  Hew,  T.  S.,  &  Hew,  J.  J.  (2018). Cloud  computing  in 
manufacturing: The next industrial revolution in Malaysia? Expert Systems with Applications, 93, 
376–394. https://doi.org/10.1016/j.eswa.2017.10.009 
18. Sanders, N. R.,  & Wagner, S. M.  (2021). Big data and supply  chain management:  A review and 
research agenda. Journal of Business Logistics, 42(1), 81–105. https://doi.org/10.1111/jbl.12264 
19. Schoenherr, T., & Speier‐Pero, C. (2015). Data science, predictive analytics, and big data in supply 
chain  management:  Current  state  and  future  potential. Journal  of Business Logistics,  36(1),  120–
132. https://doi.org/10.1111/jbl.12082 
20. Shmueli,  G., Sarstedt,  M., Hair,  J. F.,  Cheah, J.-H.,  Ting, H., Vaithilingam, S.,  & Ringle,  C.  M. 
(2019). Predictive model assessment in PLS-SEM: Guidelines for using PLSpredict. European Journal 
of Marketing, 53(11), 2322–2347. https://doi.org/10.1108/EJM-02-2019-0189 
21. Venkatesh,  V.,  Thong,  J.  Y.  L.,  &  Xu,  X.  (2016). Unified  Theory  of  Acceptance  and  Use  of 
Technology  2:  A  theoretical  model  extension. MIS  Quarterly,  40(1),  283–
301. https://doi.org/10.25300/MISQ/2016/40.1.06 
22. Wamba, S. F., Akter, S., Edwards, A., Chopin, G., & Gnanzou, D. (2020). How “big data” can make 
big impact: Findings from a systematic review and a longitudinal case study. International Journal of 
Production Economics, 165, 234–246. https://doi.org/10.1016/j.ijpe.2014.12.031 
23. Wang, G., Gunasekaran, A., Ngai, E. W. T., & Papadopoulos, T. (2022). Big data analytics in logistics 
and supply chain management: A predictive analytics perspective. Transportation Research Part E: 
Logistics and Transportation Review, 114, 210–225. https://doi.org/10.1016/j.tre.2016.04.007 
24. Wichmann, P., Brintrup, A., Baker, S., Woodall, P., & McFarlane, D. (2020). Extracting supply chain 
maps from news articles using deep neural networks. International Journal of Production Research, 
58(17), 5320–5336. https://doi.org/10.1080/00207543.2019.1671629 
25. Zailani,  S.,  Jeyaraman,  K.,  Vengadasan,  G.,  &  Premkumar,  R.  (2012). Sustainable  supply  chain 
management in Malaysia: Key drivers and performance outcomes. International Journal of Operations 
& Production Management, 32(9), 984–1011. https://doi.org/10.1108/01443571211265684 
26. Zhou,  K.,  Liu,  T.,  &  Zhou,  L.  (2020).  Industry  4.0:  Towards  future  industrial  opportunities  and 
challenges.  International  Journal  of  Production  Research,  58(6),  1922–1940. 
https://doi.org/10.1080/00207543.2019.1672905 