Federated learning (FL) and differential privacy (DP) collectively provide a robust framework for preserving privacy in healthcare data analytics. This study extends this architecture by introducing quantum-inspired gradient perturbation and examining the impact of quantum-inspired noise models and their integration order within privacy-preserving federated learning systems. A comprehensive evaluation of federated averaging (FedAvg) and federated proximal (FedProx) optimization strategies is conducted under both independent and identically distributed (IID) and non-independent and identically distributed (non-IID) data settings. A control-variate aggregator (SCAFFOLD) is also evaluated on the core non-IID setting, and the core experiments are run across three random seeds with pairwise paired t-tests for statistical significance. The findings show that all approaches achieve high classification accuracy and macro-F1 performance while preserving privacy guarantees; across seeds, noise ordering and aggregator choice produce no statistically significant difference in final accuracy (all p ≥ 0.3), confirming that the privacy-enhancing tiers preserve utility at a fixed budget. The proposed framework is further compared with classical Gaussian and Laplacian perturbation methods, and controlled ablation experiments confirm the effectiveness of the different configurations. Results indicate that quantum-inspired noise produces smoother convergence behaviour during early communication rounds while maintaining strong predictive performance. An extended evaluation on a lightweight convolutional network with two further imaging modalities (PathMNIST, OrganAMNIST), a four-qubit entangled circuit, and active privacy attacks shows that pre-DP perturbation increases gradient-inversion error by approximately 7.4× over DP-SGD alone; the budget-neutral quantum-inspired tier matches this gain at equivalent accuracy, its distinctive contribution being a statistically distinct noise distribution rather than a larger reconstruction error, while membership-inference advantage remains near chance. The evaluation includes privacy accounting, convergence stability, computational overhead, accuracy, loss, macro-F1 score, and confusion matrix analysis.
The growing demand for resilient, proactive, and intelligent supply chains (SCs) has increased the adoption of analytics to support operational, tactical, and strategic decision-making. Despite the rapid growth of analytics applications in supply chain management (SCM), many initiatives remain confined to theoretical or experimental contexts and face significant challenges in real-world implementation. Existing literature has predominantly focused on methodological advancements and performance improvements, while barriers to implementation are often treated as secondary concerns rather than structural constraints that may inhibit adoption. This study examines barriers to the adoption of analytics in SCM through a systematic literature review complemented by grey literature. Based on the analysis of 464 papers, a taxonomy of barriers was developed, comprising six categories: organizational, technological, data-related, modeling, regulatory and ethical, and people-related. The findings show that these barriers are highly interdependent and generate compatibilities and trade-offs, where improvements in one dimension may mitigate or intensify constraints in another. The integration of grey literature supported the development of a multilayered framework that conceptualizes analytics adoption as holistic and interdependent process composed of three interconnected layers: (i) a strategic layer related to alignment, direction, and value orientation; (ii) a capability layer derived from the six barrier categories and representing the organizational capabilities required to support analytics adoption; and (iii) an operationalization layer associated with process integration, scalability, and effective analytics use within SC workflows. Finally, our study outlines future research directions and managerial implications to support the adoption of analytics in SCM.
The predictability of the cryptocurrency market is a current and highly researched topic. However, the cross-market influences on the returns of the cryptocurrencies would benefit from different perspectives. This study examines the dynamic interplay between Bitcoin, Ethereum, and traditional financial indices spanning from 2017 to 2023 using the Mixture Transition Distribution and Generalized Multivariate Markov Chain models. The findings indicate the presence of own-state persistence within each cryptocurrency, as well as influence from a diverse set of financial indices. Evidence suggests that different sector indices affect these cryptocurrencies with distinct degrees of influence. Additionally, the results demonstrate cross-currency influence, notably Ethereum’s influence on Bitcoin, suggesting a more interconnected market.
This study introduces a machine learning-based framework for predicting and explaining portfolio-relative hardness in graph-based combinatorial optimization. The framework was evaluated on the maximum clique problem (MCP), where algorithmic performance can vary substantially with graph structure. Rather than treating hardness as an intrinsic property of an MCP instance, this study defines hardness empirically with respect to a heterogeneous target solver algorithms portfolio. The proposed framework consists of two stages. In the first stage, problem-agnostic graph features were computed and machine learning classification models were trained to predict whether an instance was portfolio-hard. The best-performing classifier, a support vector classifier, achieved a weighted F1-score of 0.9921, a portfolio-hard minority-class F1-score of 0.8780, and a ROC–AUC score of 0.9083 using only three graph features. In the second stage, FP-Growth association rule mining (ARM) was used to identify frequent structural signatures among portfolio-hard instances. The best association rule covered 88.29% of portfolio-hard instances, achieving a specificity of 87.63%, a balanced accuracy of 87.96%, and enriched the portfolio-hard rate by a factor of approximately 5.91 relative to the dataset base rate. Regression models were also trained to predict algorithmic computation times, with the best-performing model achieving a mean absolute percentage error of 1.7% and a coefficient of determination close to 1. Overall, the results show that graph features can support prediction of portfolio-relative solver difficulty and provide interpretable structural signatures associated with simultaneous failure across a diverse target solver portfolio.
This study proposes a framework in which financial innovation emerges through the selection of competing predictive models under capital allocation. It introduces the Principle of Epistemic Arbitrage, whereby agents adopting models with lower Kullback–Leibler (KL) divergence, a measure of how closely a model approximates the true data-generating process, attract posterior weight and capital, leading to model dominance. Within a Bayesian updating setting with log-utility investors, the analysis shows that posterior mass, wealth shares, and equilibrium prices converge to the KL-minimising model, providing an interpretation of market efficiency as capital-weighted Bayesian model selection and implying that mis-specified models generate only transient abnormal returns. The framework further demonstrates that model dominance is dynamic, as widespread adoption of a modelling framework alters the effective data-generating process, inducing endogenous model risk and the erosion of excess returns, thus offering a micro-founded perspective on model performativity where predictive tools shape the market environment in which they operate. Illustrative evidence from historical case studies and stylised empirical patterns in hedge fund survival supports the role of tail-aware models in reducing failure risk, and the study contributes to the literature on learning and market efficiency by establishing a tractable link between model selection, financial innovation, and endogenous systemic risk.
In this research, strategies for mobile robot navigation are studied and discussed critically, especially on the development of path planning methods under various environments and constraints. It discusses classical and intelligent approaches in detail to draw out their merits, demerits and scope for future research. Classical approaches investigated are Dijkstra, A-star, D-star, D-star lite, roadmap approaches and cell decomposition. Intelligent approaches studied include the particle swarm optimization algorithm, the firefly algorithm, the shuffled frog leaping algorithm, the grasshopper optimization algorithm, the bacterium foraging optimization algorithm, the bee colony optimization algorithm, the ant colony optimization algorithm, the grey wolf optimization algorithm, the red wolf optimization algorithm, and neural networks, genetic algorithms, fuzzy logic, and potential fields. Besides, the current review also includes an in-depth examination of the contemporary methods of machine-learning-based approaches, which includes the Reinforcement Learning (RL), the Deep RL, Graph Neural Networks (GNNs), and the Transformer-based methods, which occupy a dominant role in the current academic research. The review addresses navigation in both single-robot and multi-robot systems, across static and dynamic environments. The findings indicate that intelligent strategies offer greater flexibility and effectiveness in diverse terrains compared to classical methods. Moreover, intelligent approaches are often employed to enhance classical techniques through the development of hybrid algorithms. Overall, mobile robot path planning increasingly relies on intelligent approaches. The exploration is climaxed with the presentation of an intense comparative framework that includes new tables and figures that define the execution, computational complexity and flexibility of different navigation strategies. The visualizations support the relevance of the strategies in particular robotic cases, and explain the significant research paths.
Clustering remains a cornerstone of unsupervised learning, yet classical algorithms often struggle with the inherent complexities of modern datasets, such as high dimensionality, non-linear separability, and significant noise. This paper introduces the Hybrid Quantum-Classical Clustering Algorithm (HQCA), a novel framework that leverages the power of quantum computation and the robustness of classical machine learning to overcome these limitations. HQCA integrates a quantum kernel estimation phase, utilizing a parameterized quantum circuit to project data into a high-dimensional Hilbert space, thereby enhancing cluster separability where classical kernels fail. Furthermore, it employs a quantum-enhanced optimization routine, based on the Quantum Approximate Optimization Algorithm (QAOA), to determine optimal cluster centroids, thereby mitigating the common pitfall of local minima encountered by algorithms like K-Means. The algorithm also incorporates a quantum-inspired gap statistic to autonomously determine the optimal number of clusters. We rigorously evaluate HQCA against state-of-the-art classical clustering methods, including K-Means, DBSCAN, and Spectral Clustering, on multiple real-world benchmark datasets from the UCI repository, such as Iris, Wine, and a high-dimensional image dataset (Fashion-MNIST). Our results demonstrate that HQCA consistently achieves superior performance across multiple evaluation metrics, including Purity, Silhouette Score, and Adjusted Rand Index (ARI), particularly excelling in scenarios involving complex, non-linear data manifolds. The proposed hybrid paradigm not only signifies a substantial leap in clustering accuracy and robustness but also provides a scalable pathway for leveraging current Noisy Intermediate-Scale Quantum (NISQ) devices in practical data analytics workflows, marking a critical step toward potential quantum advantage in unsupervised learning. We emphasize that the demonstrated advantage is algorithmic and representational, not an asymptotic runtime advantage.
Organizations increasingly operate in complex systems that are dynamic, uncertain, data-rich, and tightly coupled, where performance must be balanced across multiple conflicting objectives and classical optimization alone often proves insufficient. Although research on the intersection of Artificial Intelligence (AI) and optimization has expanded rapidly, the literature remains fragmented across application domains and hybrid algorithm families. This paper presents a PRISMA-guided Systematic Literature Review (SLR) of 169 Scopus-indexed articles published between 2020 and January 2026. The search protocol adopts a precision-oriented Scopus strategy and supplementary eligibility screening to strengthen retrieval transparency. It examines how AI is coupled with multi-objective and goal-based optimization in managerial and operational decision contexts, the roles AI plays, the value it adds beyond classical optimization, and the decision problems for which different hybrid designs are most suitable. Combining descriptive mapping, thematic content analysis, evidence-quality interpretation, and cross-theme synthesis, the review organizes the evidence across six thematic domains, with the strongest concentration in manufacturing, transport and logistics, and energy and utilities. Four recurring AI roles emerge: predictive or representational enrichment, adaptive search, policy learning, and problem structuring. The findings show that AI creates its strongest value when classical optimization encounters four recurring bottlenecks: poor observability, combinatorial search complexity, dynamic sequential adaptation, and tightly coupled coordination. The evidence base is dominated by embedded, reinforcement-learning-intensive, multi-objective architectures, while goal-based models remain less frequent but still relevant in target-oriented and managerially interpretable settings. The review classifies major AI-optimization integration forms, links them to decision contexts, and assesses the maturity of major hybrid designs. The evidence is uneven across domains, with stronger cumulative support in manufacturing, transport, and energy than in healthcare and some service-oriented settings. While no single configuration is universally superior, future progress depends on more explainable, transferable, and implementation-ready AI-enhanced decision support systems.
Early detection and effective management of breast and leukemia cancers require robust statistical modeling to understand the influence of therapeutic and environmental factors on patient outcomes. This study introduces a novel power XGamma (PXG) lifetime model, which extends the classical XGamma distribution by incorporating an additional power parameter, thereby enabling the modeling of increasing, decreasing, and bathtub-shaped hazard rate behaviors. We investigate inferential procedures for the PXG model under adaptive Type-II progressive censoring with a binomial removal mechanism. Maximum likelihood estimators and asymptotic confidence intervals are derived for the PXG model parameters and relevant reliability and hazard functions. Bayesian inference is conducted using independent gamma and beta priors, with posterior summaries and highest posterior density intervals obtained via the Metropolis–Hastings algorithm. A comprehensive Bayesian sensitivity analysis examines the robustness of posterior estimates under informative, weak, overdispersed, and improper priors, ensuring the reliability of the inferential conclusions. Monte Carlo simulations assess estimator performance under multiple precision criteria. Applications to two real-world survival datasets on breast and blood cancers demonstrate the practical utility of the PXG model, which outperforms existing models such as the alpha-power-Weibull, power Lindley, exponentiated Weibull, and classical XGamma distributions. The PXG-based framework provides superior model fit and meaningful insights into reliability measures, while the adaptive censoring mechanism proves highly effective for timely and informative experimental termination, particularly when the primary interest lies in failure size.
We propose an original methodology for converting unstructured textual data into structured survey data through the use of chatbot technologies. Specifically, we employ ChatGPT application programming interface tools to associate a given short text – interpreted as a potential response – with the most probable question, selected from a predefined list, that could plausibly have elicited it. This list may correspond to items from a survey questionnaire. In addition, we instructed ChatGPT to identify an appropriate response option in line with the semantic content of the text. The input text may originate from (i) a post on the X platform (formerly Twitter) or (ii) an open-ended response within an actual survey questionnaire. In the former case, the method enables the construction of a structured survey dataset from social media messages; in the latter, it provides a means to assess the reliability of responses in an existing survey instrument. Our findings highlight the potential of chatbot-based methods to enhance the integration of textual and survey data, offering new opportunities for constructing timely indicators and for improving the assessment of response quality in existing survey instruments. More broadly, the approach contributes to advancing mixed-data methodologies in social research.
Decentralized online convex optimization with coupled constraints provides a natural framework for networked resource allocation, demand response, and multi-agent control. Forecasts can improve performance, but most decentralized algorithms use forecast uncertainty only as a diagnostic rather than as a control input. This paper studies a local trust-modulation mechanism that maps prediction-interval width into the weight assigned to an optimistic gradient inside a decentralized primal–dual update. Narrow intervals imply higher trust, whereas wide intervals reduce optimism and may trigger a safe fallback. For a benchmark-aligned scalar positive-part surrogate, we derive a master primal–dual inequality and resulting bounds on dynamic regret and cumulative constraint violation with explicit dependence on weighted interval width, miscoverage, comparator variation, and network mixing. Empirically, a fully causal benchmark over sixteen non-overlapping quarterly electricity-load windows from 2011 to 2014 shows that the UQ-aware controller attains lower mean dynamic regret than the ex ante causal error-adaptive safeguard (141.046 ± 22.110 versus 144.033 ± 23.141), with a mean paired regret gap of −2.987 and a mean violation gap of +3.130. Under forecast corruption, the method degrades more gracefully than the other predictive baselines, although the safe no-prediction mode becomes best under severe corruption. A post-hoc behavior-matched width-blind diagnostic narrows the incremental gap sharply, indicating that the width-specific effect beyond generic conservativeness is modest in the scalar positive-part setting studied here.
Educational Process Mining (EPM) is widely used to analyze learning processes in technology-enhanced learning environments. However, most EPM approaches remain case-centric, linking each event to a single entity (e.g., student or course), which prevents the analysis of learning situations involving multiple interacting entities, such as group collaboration or shared resource use. Recent developments in Object-Centric Process Mining (OCPM) address this limitation by allowing events to be connected to multiple related objects. Yet, the educational field lacks a practical, methodological framework for extracting structured, object-centric event data from Learning Management Systems (LMSs). Consequently, Object-Centric Educational Process Mining (OC-EPM) remains underdeveloped. This study introduces a framework for extracting multidimensional, object-centric event data from Moodle to support OC-EPM. The framework follows the OCPM2 methodology and is implemented in PM4Moodle, an open-source tool that generates OCEL-compliant logs directly from Moodle. A case study in a university course demonstrates, to our knowledge, the first application of OC-EPM in education. The results show how object-centric data enable richer analyses of interactions among students, groups, assignments, and resources, offering insights into learning processes that cannot be captured with case-centric approaches.
Project managers routinely rely on key performance indicators (KPIs) and statistical process control tools — such as Exponentially Weighted Moving Average (EWMA) and Cumulative Sum (CUSUM) — to monitor project dynamics. Yet evidence on whether structural complexity metrics improve strategic decision quality, rather than merely deviation detection, remains scarce. This paper presents an exploratory controlled between-subjects experiment evaluating the decision-quality impact of the Systemic Complexity Index (SCI), a curvature-based metric derived from Riemann–Christoffel tensor formalism, against KPI-only, KPI+EWMA, and KPI+CUSUM information conditions. Thirty-two human professionals and 8 large language models (LLMs) evaluated four decision scenarios derived from real project data (256 observations total), with decision quality operationalized through the Decision Quality Index (DQI). KPI+SCI was the only condition producing a statistically distinguishable improvement, with large effect sizes in both LLMs (d=2.716) and humans (d=1.055, ΔDQI=＋0.172 humans, ＋0.155 AI), driven exclusively by Causal/Structural Alignment and operating through a mediated pathway — systemic comprehension fully mediates the effect in humans and partially in LLMs. Fisher Z meta-analysis across five independent projects yields r=0.767 (95% CI [0.44, 0.91], I2=0%), establishing the preliminary SCI’s predictive evidence and positioning it as a potential structurally unique complement to conventional monitoring tools, offering decision-making value precisely where statistical process control methods are limited.
Just-in-time (JIT) production in flexible flow shops (FFS) is complicated by stage-dependent processing times, setup times, and machine availability, which can create utilization imbalances and lead to longer flow times. We develop a batch scheduling model to minimize total actual flow time, defined as the total time all parts spend on the shop floor from arrival to a common due date. The formulation spans multiple stages, each equipped with parallel machines. The model determines the number of batches, batch sizes, a reference stage that guides synchronization, and the schedule of the resulting batches across machines at each stage. To solve the model efficiently, we propose a Lagrangian-relaxation heuristic that yields closed-form batch-size expressions and practical procedures to select both the number of batches and the reference stage. Computational experiments consider two benchmarks: (i) a fixed reference stage with enumeration over batch counts, and (ii) a global benchmark that enumerates all combinations of reference stages and batch counts. The heuristic attains average effectiveness above 99% with small deviations and runs in O(KN), versus O(K2N2) for exhaustive search, supporting its use in large, heterogeneous JIT FFS environments.
Bicycle manufacturers face frequent supply disruptions due to long lead times, global component sourcing and complex product structures. When components are shared across product lines, a single delayed shipment can affect dozens of production orders. Practitioners typically respond with ad hoc expediting, yet have little guidance on which intervention is most effective. This paper develops and evaluates a multi-horizon framework for disruption management in bicycle manufacturing comprising three modules: (i) a planning module that adjusts already-open purchase orders 2–8 weeks ahead via mixed-integer programming, (ii) a scheduling module that swaps at-risk and safe production orders within 1–7 days and (iii) a real-time module that decides on expediting or delaying orders within 24 h using reinforcement learning. All modules share a common risk metric based on component buffer times and are coordinated through a data-driven decision table that recommends which module to apply based on urgency, exposure and schedule flexibility. Using seven months of production data from a single large European bicycle manufacturer, we evaluate the framework through an offline simulation that compares it against a reconstructed no-intervention baseline, a reactive expediting baseline and fixed single-module strategies. In this case-based simulation, the framework reduces cumulative production delay by 38.9% relative to no intervention and improves on-time completion from 67.3% to 79.6%, with a 4.7% increase in inventory holding cost and 41.2% lower intervention spending than the reactive expediting baseline. The results show that a shared risk metric and simple module-selection logic can support disruption response in a bill-of-materials (BOM)-driven manufacturing setting. Much of the benefit in this case-based simulation comes from matching responses to disruption characteristics rather than applying a single fixed intervention. The framework remains beneficial when lead-time assumptions are tightened and when scheduling soft costs are introduced, although the magnitude of the benefit narrows under stricter assumptions. Realising it in practice depends on data readiness, cross-functional governance and recalibration to firm-specific characteristics. Within these boundaries, the results provide evidence of the value of coordinating planning, scheduling and real-time responses through a shared risk signal.
The widespread adoption of Internet of Things (IoT) technologies within Industry 4.0 has increased the need for more responsive and efficient intralogistics operations, especially as industries move toward highly customized, small-batch production. Autonomous Mobile Robots (AMRs) provide the flexibility needed for dynamic industrial environments. However, their effective integration relies on accurate task-duration estimates, which are difficult to obtain due to system variability and operational uncertainty. This work addresses this challenge by statistically modeling picking and loading/unloading times using real-world industrial data. Probability distribution-fitting techniques are employed to capture the random nature of these operations, thereby reducing the need for unrealistic assumptions often used in scheduling models. The proposed methodology enhances the statistical basis for robust AMR fleet scheduling, improving coordination and reducing operational variability. Validation is conducted by directly comparing empirical data with fitted distributions, demonstrating clear improvements in representativeness and predictive accuracy. The results demonstrate the potential of this approach to improve decision-making in highly dynamic intralogistics environments.
This study examines how internal resource constraints and external institutional pressures influence the adoption of environmental, social, and governance (ESG) practices among small and medium-sized third-party logistics (3PL) firms. Drawing upon the resource-based view (RBV), institutional theory (INT), and the technology–organization–environment (TOE) framework, the research develops an interval-valued spherical fuzzy set (IVSFS) decision-support model that integrates the Criteria Importance Through Intercriteria Assessment (CIMAS), Weight by Envelope and Slope (WENSLO), and Aczel–Alsina Weighted Assessment (ALWAS) methods to prioritize ESG barriers and mitigation strategies under uncertainty. Findings reveal that limited financial capacity and the absence of public incentives are the most critical barriers, reflecting both coercive institutional pressures and internal financial constraints, while additional obstacles include fragmented regulations and organizational resistance to change. The most effective mitigation strategies involve adopting green logistics technologies, carbon accounting tools, and ESG performance dashboards, emphasizing digital innovation for enhanced transparency, eco-efficiency, and compliance. Theoretically, the study advances sustainability scholarship by integrating RBV, INT, and TOE perspectives into a unified explanatory framework. Methodologically, it contributes a hybrid multi-criteria decision-making (MCDM) model under IVSFS that strengthens uncertainty handling in sustainability assessment, and managerially, it proposes a phased ESG adoption roadmap that prioritizes cost-effective digital solutions as foundational enablers of long-term sustainability capability development.
Modern manufacturing systems operate in highly dynamic environments, where continuous innovation demands extensive monitoring and adaptation. Bottlenecks in these systems pose challenges for testing and implementing improvements without disrupting ongoing operations. Discrete Event Simulation (DES) offers a forward-looking methodology for assessing alternative scenarios within complex systems. However, the manual design of DES models is typically resource- and time-intensive, making it difficult to accommodate dynamic, rapidly changing conditions. The current approaches are largely conceptual and do not incorporate evidence-based DES model building. This limits their effectiveness and scalability, particularly in modern manufacturing environments characterized by high variability and constant evolution. This study targets the complexity of manually designing DES environments by proposing an automated framework—WEFTSIM—that extracts simulation models directly from manufacturing data. The goal is to enhance the accuracy, efficiency, and adaptability of DES models in representing real-world processes. The evaluation combines two approaches: (i) A real-world case-study-driven assessment of WEFTSIM’s applicability to the automatic, data-driven design of DES models through a systematic comparison of scenario performance. (ii) Validation assessment through a comparative performance analysis of the proposed methods against historical data to quantify their fidelity in representing the system behavior. WEFTSIM effectively derived DES models that closely mirror the actual manufacturing operations, attaining around 80% coverage at the activity level and achieving a high trace similarity of 90% when validated against the observed data. The automated approach reduced the manual and time-intensive conceptualization phase. The evaluation against existing benchmark methods shows WEFTSIM’s capability to automatically design DES models, detect bottlenecks, and rapidly identify an improvement scenario.
Strategic projects are exposed to multiple uncertainties within a competitive environment that result in trade-off between two decisions, early investment that pre-empts competitors versus delay to allow resolution of uncertainties. Unlike the binary net present value (NPV) based decision to invest or abandon now, our multiple volatility real option games approach (MVR games) frames strategic investment decisions as a dynamic process for future options. It is a decision tree framework that computes and analyses strategic enhanced net present value (SNPV) using four MVR games modules and decision making. SNPV combines NPV with two offsetting real option and strategic values. Real option value arises from flexibility in investment decisions responding to random uncertainties. Strategic value reduces it due to uncertainty from competitive interactions between rival firms. Module 1 performs Monte Carlo simulations to obtain volatility of project value related to each risk variable. Module 2 uses multiple volatilities to build MVR games tree with project values following binomial process. When the tree terminates on option expiry, Nash equilibrium payoffs and predicted investment decision outcome (IDO) probabilities are obtained. The tree rolls back to obtain SNPV at the root to decide whether to hold the option till expiry or not. Module 3 demonstrates SNPV convergence to stable SNPV and its relationship with IDO probabilities. Module 4 conducts robust analysis of SNPV and IDO probabilities. MVR games decision to invest or abandon project includes decision problems, objective functions, trade-offs, alternative actions and preferences. MVR games are applied to real-world projects for strategic investment decisions by rival firms.
The framework of disruptive innovation has changed how we analyse and interpret many successful businesses, especially those that start from a disadvantageous position. A wealth of real-world cases has been studied to offer valuable insights into what disruptive innovation is and how it evolves. However, case studies only focus on a small set of numerous possible outcomes, unable to answer “what-if” questions – essential for revealing the core factors/mechanisms that define disruptive innovation. Following the philosophy of generative social science, this research builds an analytics framework that incorporates an agent-based model, aiming to “grow” typical diffusion patterns of disruptive innovation from microscopic consumer decisions while investigating underexplored mechanisms and alternative outcomes via numerical experiments. The duality of value dimensions is assumed to be the most essential characteristic of disruptive innovation in the framework. Other factors, including heterogeneous consumer preferences, complementary technologies, technological progress, and pricing, are experimented with as key factors that may shape the diffusion processes. The results demonstrate that the model can reproduce multiple diffusion patterns and stylised facts regarding disruptive innovation, such as market encroachment on low-end markets, opening new markets, capturing market shares from mainstream markets, being impacted by new disruptors, and its coexistence with sustaining innovation. The analysis indicates that the dual value dimensions endow an innovation with the potential of being disruptive, while other factors can adjust its actual disruptive effect. Based on the findings, theoretical and practical managerial suggestions are provided for disruptive innovation researchers and practitioners.
