RARick Acosta-Vega

Rick Acosta-Vega

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Rick Acosta-Vega has a total of 24 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Evaluation of Machine-Learning-Guided Adaptive Sampling for Industrial Monitoring Systems

    by Jose J. Muñoz, Manuel J. Campuzano, Rick Acosta-Vega

    Machine learning (ML) can guide measurement allocation, but predictive accuracy does not establish operational value. We evaluate whether a bounded ML allocator improves industrial monitoring when a frozen variance detector retains alarm authority. The evaluation combines five run-grouped test folds on the Skoltech Anomaly Benchmark (SKAB), an illustrative two-event MetroPT-3 replay, and 1800 controlled policy runs. On SKAB, ML rollout achieved physical-event recall of 0.636 versus 0.697 for the smallest fixed policy; the paired 95% interval for the difference included zero. It required 5.985 additional observations per system epoch, an acquisition proxy rather than measured energy or monetary expenditure. Precursor skill varied across runs, and controlled tests exposed sensitivity to distribution shift. Delay-only break-even analysis identified a high cost ratio needed to favor ML. MetroPT-3 illustrates cadence sensitivity but cannot establish field reliability. The contribution is a reproducible seven-gate evaluation procedure linking precursor validity, detector suitability, resource use, uncertainty, and deployment decisions. The findings support withholding autonomous deployment of this specific univariate variance-monitoring configuration; they do not establish the performance of multivariate or distributionally robust allocation methods.

  • Open Access

    Resource-Constrained Adaptive Sampling for Variance Monitoring: Comparing Rule-Based, MLP, and LSTM Acquisition Policies

    by Manuel J. Campuzano, Rick Acosta-Vega, Natalia Chaparro-Hernández

    Automated monitoring must balance detection delay and measurement use. This study compares rule-based, feedforward multilayer perceptron (MLP), and long short-term memory (LSTM) acquisition policies coupled to a common exponentially weighted moving average (EWMA) variance detector. Experiments 1 and 2 comprise 714,000 and 660,000 test trajectories, respectively. Closely matched rule and LSTM configurations achieve in-control average run lengths of 369.97 and 370.11, consumption of 4.918 and 4.917 observations per epoch, and weighted detection delays of 18.860 and 18.864 epochs; the conditional 95% bootstrap confidence interval for the difference in their weighted delays includes zero. A third experiment repeats optimization, selection, calibration, and testing ten times per family, comparing a search budget of 512 candidate evaluations with one of 30 CPU seconds. The mean LSTM-minus-rule delay contrasts are −0.219 and +0.206 epochs, respectively, with across-run 95% intervals [−0.723, 0.285] and [−0.295, 0.708]. Scenario weights, measurement prices, and acquisition duration further affect policy comparisons. The findings characterize the evaluated finite search procedures and operating points within a common detector; they establish neither intrinsic superiority nor equivalence of the policy architectures.

  • Open Access

    Trajectory-Calibrated Multivariate Control Charts with PCA and Machine Learning

    by Manuel J. Campuzano, Rick Acosta-Vega, Enrique Delahoz-Domínguez

    Multivariate control charts combine a monitoring statistic, a control limit, and a sequential signaling rule. This study evaluates the incremental contribution of principal component analysis (PCA) features to a learned score chart based on lagged measurements. Each chart is calibrated using independent healthy trajectories and a rule requiring three consecutive exceedances, under a common nominal run-level alarm criterion. The charts are compared in two simulation case studies—an eight-channel autoregressive (AR) benchmark with ten repeated fitting experiments and a Tennessee Eastman Process (TEP) dataset—using healthy-run alarm rates, detection proportions, and a mean capped signal delay. This delay criterion is measured in post-change samples and penalizes faults undetected within the monitoring horizon. In the AR case, adding residual energy Q to the model using raw lags reduced the mean capped signal delay from 45.0 to 41.2 samples; the full PCA–ML and multivariate exponentially weighted moving average (MEWMA) charts had mean capped delays of 41.5 and 36.1 samples, respectively. In the TEP case, excluding three provider-flagged faults, mean capped delays were 78.6 samples for raw-lag learning, 27.2 for lags plus Q, 29.5 for full PCA–ML, and 16.8 for MEWMA. Realized healthy-run alarm rates differed across procedures. Family-level analyses relate the AR results to fault-direction geometry. Measurement noise increased alarm rates, and covariance regularization affected the TEP MEWMA response to noise. The findings support a comparator-specific benefit from residual features, without establishing general superiority, detection of fault mechanisms absent from supervised training, or performance on operating-plant data.

  • Open Access

    Adaptive and Explainable Hybrid Statistical–Machine Learning Framework for Intelligent Industrial Process Monitoring

    by Manuel J. Campuzano, Rick Acosta-Vega, Néstor Caicedo-Solano

    An industrial monitor that changes its evidence weights also changes the rule that must be calibrated and explained. We combine principal-component residual energy, latent-space multivariate exponentially weighted moving averages, and a boosted-tree corruption detector in a fixed-reference monitoring policy. Past evidence determines the weights, complete episodes calibrate the persistent alarm rule, and an additive audit reconstructs each decision. Eight independently fitted simulation experiments give mean capped delays of 158.3 samples for hybrid adaptation, 180.8 for equal-weight fusion, and 147.5 for tuned static fusion. Controlled substitutions preserve the channel count and weight path: adding learned evidence increases delay by 15.7 samples at a common limit, but reduces it by 15.3 after separate calibration. The benefit therefore depends on the interaction between evidence and calibration. The evaluation includes isolation-forest comparators and explanation tests across nine scenarios. Stable attributions do not consistently identify perturbed variables. Continuous DAMADICS replay produces seven clean signals during 16 documented interventions and 27.2% background alarm activity. Actuator-2 distribution shifts persist within coarse controller-output strata, but their operational cause remains unknown. The framework supports auditable decisions and finite-episode calibration under exchangeability; it does not establish physical diagnosis or continuous-stream alarm control.

  • Open Access

    Class-Conditional Allocation of Forecasting Models for Retail Inventory Replenishment: Evidence from an ABC-Segmented Hardware Portfolio

    by Gina Marcela Hernández-Orozco, Juan David Reales-Barragán, Verónica Deibe-Blanco, Rick Acosta-Vega, Manuel J. Campuzano

    Background: Retail inventories are heterogeneous, yet demand forecasting studies typically select one model for an entire portfolio, leaving open how forecasting effort should be allocated across inventory segments. Methods: Six models—ARIMA, SARIMA, ARIMAX, Random Forest, LightGBM and an LSTM network—were estimated on weekly demand for a Colombian hardware retailer (35,399 transactions, 1633 SKUs, January 2025–March 2026). Because 95.5% of items are intermittent or lumpy, models were fitted to three ABC-class-aggregated series of 60 observations, evaluated on a 12-week holdout and under a rolling-origin scheme, and compared against naive benchmarks using MASE and RMSSE. Results: The best model differs by class. In Class A a statistical specification wins, ARIMAX on the holdout (RMSE 843 units per week) but ARIMA under rolling origins, the two being statistically indistinguishable. Random Forest is best in Class B (255). In Class C no model improves on a last-value rule (75 against 66). Relative to that benchmark, error falls to 39.8% in Class A and 47.6% in Class B. Diebold–Mariano tests separate 16 of 45 pairs on the holdout and three of 30 under rolling origins. Conclusions: Allocating models by inventory class is operationally justified rather than statistically decisive, and the low-turnover class warrants no model at all.

  • Open Access

    Integrating Absorbing Markov Chains and Multi-Issue Bankruptcy Problems with Cross-Claims for Resource Allocation Under Production Uncertainty

    by Rick Acosta-Vega, Samuel Alvarez-Cayón, Manuel J. Campuzano

    Production systems may generate resource scarcity as products progress through production stages that are subject to deterioration, reprocessing, or failure. This study builds upon the integration of absorbing Markov chains with Multi-Issue Bankruptcy Problems with Cross-Claims (MI-MIBC) and develops a method for deriving a multi-resource estate based on the fundamental matrix of a Markov chain and state-specific multi-resource requirements. This method defines a multi-resource allocation problem with connected claims. The resulting MI-MIBC is addressed using Constrained Proportional Awards (CPA), Constrained Equal Awards (CEA), and Constrained Sequential Priority (CSP) rules. This method is illustrated with the postharvest system of Colombian plantains, which consists of money, labor, transportation, and packing resources. The three rules produce different allocation patterns when scarcity is identical. The cross-claims structure may also leave a residual estate even when claims remain unmet. This method interlinks stochastic production systems with multi-resource allocation and derives the estate from the stochastic production model rather than specifying it independently at the allocation stage.

  • Open Access

    Marine Protected Area Research and Global Governance Priorities: Thematic Shifts, Geographic Inequalities, and Science–Policy Alignment

    by Javier De La Hoz-M, Humberto Llinás, Manuel J. Campuzano, Enrique Delahoz-Domínguez, Rick Acosta-Vega

    Marine Protected Areas (MPAs) are central to global ocean conservation, yet the latent thematic organization of three decades of MPA research has not been systematically characterized. This study applies BERTopic, a transformer-based neural topic modeling framework, to 10,735 peer-reviewed publications indexed in Scopus and Web of Science (1992–2025). Eight semantically coherent research themes were identified, and their structural relationships were quantified using the Relative Importance Factor (RIF) Index, a rank-based dominance measure derived from discrete power-law modeling. Marine Conservation Governance and Marine Biodiversity Patterns together account for 64.3% of the scientific output, forming the cognitive core of the field. Temporal analysis reveals that marine pollution, megafauna movement ecology, and Chinese coastal governance are the fastest-growing themes, yet they remain structurally peripheral, a phenomenon described here as accelerated marginality. Geographic analysis further identifies a pronounced epistemic governance gap, whereby countries hosting many of the world’s most valuable marine ecosystems contribute disproportionately little to the knowledge base guiding global conservation policy. Taken together, these findings carry a policy implication: achieving the objectives of the Kunming–Montreal Global Biodiversity Framework will require not only expanding protected areas, but also rebalancing the thematic and geographic architecture of the research that underpins them.

  • Open Access

    Game Theory in Artificial Intelligence and Machine Learning: An Integrated Computational Framework Based on Bibliometrics, Topic Modeling, and HJ-Biplot

    by Juan Reales-Barragán, Javier De La Hoz-Maestre, Rick Acosta-Vega

    Game Theory provides a foundation for multi-agent systems, reinforcement learning, mechanism design, and adversarial learning within AI and ML. Nevertheless, few comprehensive studies map the structure and branches of this cross-disciplinary field. The current research attempts to fill this gap by providing a combined bibliometric and semantic study of 6974 records obtained from Scopus and Web of Science during the years 1972 to 2025. This research is among the first to implement traditional bibliometrics integrated with LDA (Latent Dirichlet Allocation) topic modeling through R (version 4.5.1) to map research networks and surface latent research themes across domains such as multi-agent reinforcement learning and adversarial learning. The major contribution is the development of an evolutionary model that identifies the Game Theory’s paradigmatic evolution in the context of AI and ML from 1972 to 2025. This research also provided the characteristics that describe the shift from the rational equilibrium paradigms to intelligent systems that learn, collaborate, and devise strategies autonomously.

  • Open Access

    Mapping AI and Data-Driven Research in Financial and Risk Analytics Using LDA and HJ-Biplot: A Bibliometric-Computational Framework

    by Lois Soto, Sofia Romero, Rick Acosta-Vega

    The rapid convergence of artificial intelligence, machine learning, and financial and risk analytics has produced a large and fragmented body of scientific literature, making it difficult to identify its thematic structure and evolutionary trends. This study addresses that gap by proposing a bibliometric-computational framework that integrates Latent Dirichlet Allocation (LDA) and the HJ-Biplot to map research at the intersection of artificial intelligence and data-driven methods with financial and risk analytics. Bibliographic records were retrieved from Scopus and Web of Science, yielding a corpus of 20,213 documents published between 1999 and 2025 from 4640 sources. After preprocessing, LDA was applied to extract latent topics, and the HJ-Biplot was used to represent the relationships among topics, years, countries, and scientific sources within a multivariate framework. The number of topics was selected through a grid search over every integer value of K between 5 and 40, using probabilistic topic coherence together with manual inspection of topic separability. The analysis identified 36 latent topics (K = 36) spanning portfolio optimization, risk assessment, fintech, and deep learning-based forecasting, and shows a rapid post-2010 expansion in the topics concerned with artificial intelligence and predictive analytics, with output concentrated in China, the United States, and India. Increasing, declining, and fluctuating topics were distinguished by the sign and statistical significance of a linear trend in annual topic prevalence. By combining probabilistic topic modeling with multivariate biplot representation, this study offers a more granular and data-driven alternative to conventional bibliometric counting methods, providing a replicable framework for tracking the evolution of interdisciplinary research fields. The design is descriptive: it characterizes the structure of the indexed literature and does not measure financial practice or establish the causes of the patterns reported.

  • Open Access

    Effective Cost Allocation in Agricultural Production Systems Using Absorbing Markov Chains and Bankruptcy Rules

    by Rick Acosta-Vega, Manuel J. Campuzano, Samuel Alvarez-Cayón

    Agricultural production systems are frequently affected by uncertainty, processing cycles, quality variability, and irreversible losses, leading to discrepancies between planned and actually available operational resources. This study proposes an integrated framework combining absorbing Markov chains and bankruptcy allocation theory to address endogenous scarcity in stochastic production systems. The production process is modeled using an absorbing Markov chain, where transient states represent operational stages and absorbing states represent successful completion or irreversible loss. The fundamental matrix is used to estimate the effective expected resource availability induced by the stochastic dynamics of the system. When this availability is insufficient to satisfy nominal operational requirements, the problem is reformulated as a bankruptcy allocation problem. Four classical allocation rules are evaluated and compared: proportional, constrained equal awards, constrained equal losses, and the Talmud rule. A stylized cocoa production system illustrates the proposed framework. The results show that the constrained equal awards and Talmud rules better preserve operational continuity under scarcity conditions. The main contribution lies in linking stochastic production modeling with normative resource allocation under endogenous scarcity.