Alessandro Albano, Chiara Di Maria, Mariangela Sciandra, Antonella Plaia
Textual analysis has gained significant interest in medical research, particularly for automated patient diagnosis based on clinical narratives. While traditional approaches often focus on associational methods, this paper explores the application of causal forests to analyze textual data from electronic health records (EHRs), aiming to identify causal relationships between specific words and the likelihood of receiving certain medical diagnoses. Utilizing the MIMIC-III dataset, we assess how linguistic factors influence diagnosis probabilities for three conditions: diabetes, hypothyroidism, and adrenal gland disorders. Our findings reveal significant causal links between certain clinical terms and diagnosis probabilities, emphasizing the potential of causal inference techniques to improve the analysis of language in clinical narratives. Additionally, we uncover heterogeneity in treatment effects, demonstrating that specific words can identify high-risk patient subgroups. This study highlights the importance of integrating causal inference in natural language processing within healthcare settings.
{"title":"Causal Forests for Discovering Diagnostic Language in Electronic Health Records","authors":"Alessandro Albano, Chiara Di Maria, Mariangela Sciandra, Antonella Plaia","doi":"10.1002/asmb.70038","DOIUrl":"https://doi.org/10.1002/asmb.70038","url":null,"abstract":"<p>Textual analysis has gained significant interest in medical research, particularly for automated patient diagnosis based on clinical narratives. While traditional approaches often focus on associational methods, this paper explores the application of causal forests to analyze textual data from electronic health records (EHRs), aiming to identify causal relationships between specific words and the likelihood of receiving certain medical diagnoses. Utilizing the MIMIC-III dataset, we assess how linguistic factors influence diagnosis probabilities for three conditions: diabetes, hypothyroidism, and adrenal gland disorders. Our findings reveal significant causal links between certain clinical terms and diagnosis probabilities, emphasizing the potential of causal inference techniques to improve the analysis of language in clinical narratives. Additionally, we uncover heterogeneity in treatment effects, demonstrating that specific words can identify high-risk patient subgroups. This study highlights the importance of integrating causal inference in natural language processing within healthcare settings.</p>","PeriodicalId":55495,"journal":{"name":"Applied Stochastic Models in Business and Industry","volume":"41 5","pages":""},"PeriodicalIF":1.5,"publicationDate":"2025-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/asmb.70038","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144897465","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
During the manufacturing processes for the integrated circuit (IC) products, defective units may not be screened out by the quality inspections. The defective units often lead to infant mortality failure in the early stages of operation, while non-defective units will eventually fail due to wear-out failure. The general limited failure population (GLFP) model can be used to describe such a phenomenon in which defective units induce failure affected by both failure mechanisms, but failure of non-defective units is only due to wear-out. Besides, when a failure occurs, it is not known whether it is defective and yet which failure mode causes the failure. This article proposes an EM algorithm along with the missing information principle for the GLFP models under multiply censored Weibull distributions to simplify the maximum likelihood (ML) inference. It resolves the computational instability and provides more accurate reliability inference. With the embedded latent variables, failure mode detection and defect identification are also made for masked data, consequently. Furthermore, the proposed method can be extended to the GLFP models of interval data. The simulation study shows that the proposed method provides more accurate results. Two illustrative examples highlight the feasibility and advantages of the proposed approach.
{"title":"Reliability Inference in GLFP Models Based on EM Algorithm With Related Application","authors":"Chih-Ying Tai, Tsai-Hung Fan","doi":"10.1002/asmb.70030","DOIUrl":"https://doi.org/10.1002/asmb.70030","url":null,"abstract":"<div>\u0000 \u0000 <p>During the manufacturing processes for the integrated circuit (IC) products, defective units may not be screened out by the quality inspections. The defective units often lead to infant mortality failure in the early stages of operation, while non-defective units will eventually fail due to wear-out failure. The general limited failure population (GLFP) model can be used to describe such a phenomenon in which defective units induce failure affected by both failure mechanisms, but failure of non-defective units is only due to wear-out. Besides, when a failure occurs, it is not known whether it is defective and yet which failure mode causes the failure. This article proposes an EM algorithm along with the missing information principle for the GLFP models under multiply censored Weibull distributions to simplify the maximum likelihood (ML) inference. It resolves the computational instability and provides more accurate reliability inference. With the embedded latent variables, failure mode detection and defect identification are also made for masked data, consequently. Furthermore, the proposed method can be extended to the GLFP models of interval data. The simulation study shows that the proposed method provides more accurate results. Two illustrative examples highlight the feasibility and advantages of the proposed approach.</p>\u0000 </div>","PeriodicalId":55495,"journal":{"name":"Applied Stochastic Models in Business and Industry","volume":"41 4","pages":""},"PeriodicalIF":1.5,"publicationDate":"2025-08-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144767709","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The complexity and dynamic nature of cyber risks pose considerable challenges to risk management. From an actuarial perspective, we propose an advanced aggregate loss process using a variant of the Hawkes process as its frequency model. The refined Hawkes process first considers the impact of loss magnitude on the frequency of risk occurrences by integrating the loss covariate into the conditional intensity function. Second, we employ a more flexible kernel function in place of the classical exponential case. By incorporating the concept of age-dependent population structure, we calculate the probabilistic properties (mean, variance) for the proposed aggregate loss process. Furthermore, numerical simulations for cyber insurance pricing are conducted based on two pricing principles. Finally, we verify the feasibility of the proposed model based on a publicly available cyber breach data set. Considering the complex and dynamic nature of cyber risks, the efficiency of the proposed model is still limited by some factors, such as the authenticity and accuracy of the data. These are worthy of further consideration in future studies.
{"title":"The Modeling of Cyber Risk Insurance by Hawkes Processes With Loss Covariate","authors":"Na Ren, Xin Zhang","doi":"10.1002/asmb.70026","DOIUrl":"https://doi.org/10.1002/asmb.70026","url":null,"abstract":"<div>\u0000 \u0000 <p>The complexity and dynamic nature of cyber risks pose considerable challenges to risk management. From an actuarial perspective, we propose an advanced aggregate loss process using a variant of the Hawkes process as its frequency model. The refined Hawkes process first considers the impact of loss magnitude on the frequency of risk occurrences by integrating the loss covariate into the conditional intensity function. Second, we employ a more flexible kernel function in place of the classical exponential case. By incorporating the concept of age-dependent population structure, we calculate the probabilistic properties (mean, variance) for the proposed aggregate loss process. Furthermore, numerical simulations for cyber insurance pricing are conducted based on two pricing principles. Finally, we verify the feasibility of the proposed model based on a publicly available cyber breach data set. Considering the complex and dynamic nature of cyber risks, the efficiency of the proposed model is still limited by some factors, such as the authenticity and accuracy of the data. These are worthy of further consideration in future studies.</p>\u0000 </div>","PeriodicalId":55495,"journal":{"name":"Applied Stochastic Models in Business and Industry","volume":"41 4","pages":""},"PeriodicalIF":1.5,"publicationDate":"2025-08-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144767710","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Paul Pao-Yen Wu, Yu Yi Yu, Liam A. Toohey, Michael Drew, Scott A. Sisson, Clara Grazian, Kerrie Mengersen
<p>We greatly appreciate the commentary and positive feedback of discussants Prof. Jialiang Li and Dr. Rhythm Grover to enrich our paper and its context.</p><p>As noted by Prof. Li, survival models are highly applicable to the subsequent sports injury problem given the temporal dimension of injury data. In the sporting context, censoring can arise, for example, from finite surveillance windows associated with a sporting season, athletes joining and leaving a team, or even extended absence due to injury [<span>1, 2</span>]. However, given the complex systems nature of individual athletes and potentially changing dynamics and susceptibility to injury over time, it is also important to capture the changing state of the athlete explicitly [<span>3</span>]. For example, increasing strength with training over a season could reduce injury risk; however, a serious injury such as an ACL injury could lead to increased susceptibility to subsequent injuries.</p><p>Our paper presented a pragmatic approach, as noted by Dr. Grover, to tackle the challenges of modeling subsequent injury, reducing dimensionality through a time-varying Cox Proportional Hazards (PH) model, and using a discrete-time HMM to capture changes in susceptibility and covariate effects over time. Both Prof. Li and Dr. Grover note the potential computational challenge associated with Hidden Markov Models (HMMs) especially in the presence of large-scale and high-dimensional datasets. Hence, the need for dimension reduction, which was undertaken using survival modeling to explicitly cater for the time-to-event nature of injury data and censoring. The appropriateness of using the survival model was supported by checks of the assumptions of the PH model (e.g., proportional hazards, Schoenfeld residuals) and validation results (concordance index) as reported in our paper.</p><p>In addition to computational complexity, however, is the somewhat associated challenge of model convergence. Greater model complexity, such as more HMM states or more model covariates, can lead to challenges with model identifiability, estimation, computation, and thus model convergence [<span>4</span>]. This is a current research challenge when faced with limited data as in our subsequent injury application, which is limited to 33 players and 2523 training and competition sessions over one season. Computationally, the proposed discrete-time HMM fitted with Expectation Maximization (EM) took approximately 155 s to converge for the entire team of players over one season, compared to less than a second for the Cox PH model. However, model convergence with more than two states could not be achieved with this limited dataset. Therefore, although the computational cost is feasible in this case study, the data available can limit the level of model complexity that can be achieved. Hence, it highlights the utility of the proposed combination of dimension reduction and state space modelling as a more generalizable approach, and th
{"title":"Rejoinder to Next Generation Models for Subsequent Sports Injuries by Wu et al.","authors":"Paul Pao-Yen Wu, Yu Yi Yu, Liam A. Toohey, Michael Drew, Scott A. Sisson, Clara Grazian, Kerrie Mengersen","doi":"10.1002/asmb.70035","DOIUrl":"https://doi.org/10.1002/asmb.70035","url":null,"abstract":"<p>We greatly appreciate the commentary and positive feedback of discussants Prof. Jialiang Li and Dr. Rhythm Grover to enrich our paper and its context.</p><p>As noted by Prof. Li, survival models are highly applicable to the subsequent sports injury problem given the temporal dimension of injury data. In the sporting context, censoring can arise, for example, from finite surveillance windows associated with a sporting season, athletes joining and leaving a team, or even extended absence due to injury [<span>1, 2</span>]. However, given the complex systems nature of individual athletes and potentially changing dynamics and susceptibility to injury over time, it is also important to capture the changing state of the athlete explicitly [<span>3</span>]. For example, increasing strength with training over a season could reduce injury risk; however, a serious injury such as an ACL injury could lead to increased susceptibility to subsequent injuries.</p><p>Our paper presented a pragmatic approach, as noted by Dr. Grover, to tackle the challenges of modeling subsequent injury, reducing dimensionality through a time-varying Cox Proportional Hazards (PH) model, and using a discrete-time HMM to capture changes in susceptibility and covariate effects over time. Both Prof. Li and Dr. Grover note the potential computational challenge associated with Hidden Markov Models (HMMs) especially in the presence of large-scale and high-dimensional datasets. Hence, the need for dimension reduction, which was undertaken using survival modeling to explicitly cater for the time-to-event nature of injury data and censoring. The appropriateness of using the survival model was supported by checks of the assumptions of the PH model (e.g., proportional hazards, Schoenfeld residuals) and validation results (concordance index) as reported in our paper.</p><p>In addition to computational complexity, however, is the somewhat associated challenge of model convergence. Greater model complexity, such as more HMM states or more model covariates, can lead to challenges with model identifiability, estimation, computation, and thus model convergence [<span>4</span>]. This is a current research challenge when faced with limited data as in our subsequent injury application, which is limited to 33 players and 2523 training and competition sessions over one season. Computationally, the proposed discrete-time HMM fitted with Expectation Maximization (EM) took approximately 155 s to converge for the entire team of players over one season, compared to less than a second for the Cox PH model. However, model convergence with more than two states could not be achieved with this limited dataset. Therefore, although the computational cost is feasible in this case study, the data available can limit the level of model complexity that can be achieved. Hence, it highlights the utility of the proposed combination of dimension reduction and state space modelling as a more generalizable approach, and th","PeriodicalId":55495,"journal":{"name":"Applied Stochastic Models in Business and Industry","volume":"41 4","pages":""},"PeriodicalIF":1.5,"publicationDate":"2025-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/asmb.70035","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144725648","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Consider a non-standard renewal risk model in which claims arrive in pairs and the stochastic discounting process is given by , where is a Lévy process. We are interested in the joint tail probability of and
Jacob Majakwara, Patrick L. Mthisi, Honest W. Chipoyera
Credit risk management plays a crucial role in financial institutions by identifying, assessing and controlling the credit risks arising from lending activities. However, missing data pose a common problem in credit risk modelling, leading to biased estimates and a loss of statistical power. To address this issue and improve predictive accuracy, multiple imputation methods are increasingly employed. This study evaluates the performance of the Multivariate Imputation by Chained Equations (MICE) method in identifying factors associated with time to default, using the publicly available Prosper personal loan data. The analysis is conducted within the framework of mixture cure rate models based on the generalised gamma family of distributions. This research is the first of its kind to integrate the MICE approach into mixture cure rate modelling. The flexibility of the generalised gamma distribution was utilised to select the optimal mixture cure rate model. The estimated cure rate using complete cases (CC) was higher than that obtained using MICE imputation. This highlights the potential pitfalls of solely relying on CC analysis in survival analysis.
{"title":"Enhancing Credit Risk Management Through Integration of Multiple Imputation Methodology and Long-Term Survival Modelling","authors":"Jacob Majakwara, Patrick L. Mthisi, Honest W. Chipoyera","doi":"10.1002/asmb.70027","DOIUrl":"https://doi.org/10.1002/asmb.70027","url":null,"abstract":"<p>Credit risk management plays a crucial role in financial institutions by identifying, assessing and controlling the credit risks arising from lending activities. However, missing data pose a common problem in credit risk modelling, leading to biased estimates and a loss of statistical power. To address this issue and improve predictive accuracy, multiple imputation methods are increasingly employed. This study evaluates the performance of the Multivariate Imputation by Chained Equations (MICE) method in identifying factors associated with time to default, using the publicly available Prosper personal loan data. The analysis is conducted within the framework of mixture cure rate models based on the generalised gamma family of distributions. This research is the first of its kind to integrate the MICE approach into mixture cure rate modelling. The flexibility of the generalised gamma distribution was utilised to select the optimal mixture cure rate model. The estimated cure rate using complete cases (CC) was higher than that obtained using MICE imputation. This highlights the potential pitfalls of solely relying on CC analysis in survival analysis.</p>","PeriodicalId":55495,"journal":{"name":"Applied Stochastic Models in Business and Industry","volume":"41 4","pages":""},"PeriodicalIF":1.3,"publicationDate":"2025-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/asmb.70027","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144673029","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Wilmer Martínez-Rivera, Eliana González-Molano, Edgar Caicedo-Garcia
We forecast inflation aggregates for the United States, the United Kingdom, and Colombia using forecasts aggregation of disaggregates and forecasts obtained directly from the aggregate. We implement helpful models for many predictors, such as dimension reduction, shrinkage methods, machine learning models, and traditional time-series models (ARIMA and TAR). We evaluate out-sample forecasts for the period before COVID-19 and the period afterward. It was found that the aggregation of forecasts performs as well as the forecast using the aggregate directly. In some cases, there is a reduction in the forecast error from the disaggregate analysis.
{"title":"Forecasting Inflation From Disaggregated Data","authors":"Wilmer Martínez-Rivera, Eliana González-Molano, Edgar Caicedo-Garcia","doi":"10.1002/asmb.70023","DOIUrl":"https://doi.org/10.1002/asmb.70023","url":null,"abstract":"<div>\u0000 \u0000 <p>We forecast inflation aggregates for the United States, the United Kingdom, and Colombia using forecasts aggregation of disaggregates and forecasts obtained directly from the aggregate. We implement helpful models for many predictors, such as dimension reduction, shrinkage methods, machine learning models, and traditional time-series models (ARIMA and TAR). We evaluate out-sample forecasts for the period before COVID-19 and the period afterward. It was found that the aggregation of forecasts performs as well as the forecast using the aggregate directly. In some cases, there is a reduction in the forecast error from the disaggregate analysis.</p>\u0000 </div>","PeriodicalId":55495,"journal":{"name":"Applied Stochastic Models in Business and Industry","volume":"41 4","pages":""},"PeriodicalIF":1.3,"publicationDate":"2025-07-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144573576","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
If Francis Bacon were born today, he might have said “data is power” instead of his original saying, “knowledge is power.” In modern society, data is everywhere. In memory of Deming (a guru in quality), this paper attempts to address the fundamental issue of data quality and how Deming would handle it. Specifically, we attempt to explain what data quality really means, and the critical impact that it has on data science. Statisticians, who understand how to collect high quality data, have much more to contribute to both the intellectual vitality and the practical utility of data science. At the same time, data science challenges statisticians to move out of some familiar habits to engage less structured problems, to become more comfortable with ambiguity, and to engage more scientists in a fruitful discussion on what various parties can bring to this new mode of investigation. Some potential avenues for future research in the collection of high-quality data will be proposed.
{"title":"Data Quality: What if Deming Were Born Today?","authors":"Dennis K. J. Lin, Nicholas Rios","doi":"10.1002/asmb.70025","DOIUrl":"https://doi.org/10.1002/asmb.70025","url":null,"abstract":"<p>If Francis Bacon were born today, he might have said “data is power” instead of his original saying, “knowledge is power.” In modern society, data is everywhere. In memory of Deming (a guru in quality), this paper attempts to address the fundamental issue of data quality and how Deming would handle it. Specifically, we attempt to explain what data quality really means, and the critical impact that it has on data science. Statisticians, who understand how to collect high quality data, have much more to contribute to both the intellectual vitality and the practical utility of data science. At the same time, data science challenges statisticians to move out of some familiar habits to engage less structured problems, to become more comfortable with ambiguity, and to engage more scientists in a fruitful discussion on what various parties can bring to this new mode of investigation. Some potential avenues for future research in the collection of high-quality data will be proposed.</p>","PeriodicalId":55495,"journal":{"name":"Applied Stochastic Models in Business and Industry","volume":"41 4","pages":""},"PeriodicalIF":1.3,"publicationDate":"2025-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/asmb.70025","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144514730","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
<p>This study introduces a statistical methodology for document clustering that integrates multiple dimensions of textual similarity through network topology analysis. The proposed methodology, which we call Multi-dimensional Similarity Network Analysis (MSNA), extends traditional document-clustering approaches by combining semantic embeddings, topic probability distributions, and emotional probability distribution into a unified similarity measure. We formalize this through a weighted combination of Jensen-Shannon divergences across different probability spaces, creating a comprehensive similarity network. The clustering is achieved through a community detection algorithm that optimizes a multi-objective modularity function, accounting for the different similarity dimensions. We prove the statistical consistency of our approach and derive bounds for the clustering performance under mild regularity conditions. The methodology is validated on a large-scale data set of Airbnb reviews <span></span><math> <semantics> <mrow> <mo>(</mo> <mi>n</mi> <mo>=</mo> <mn>114</mn> <mo>,</mo> <mn>000</mn> <mo>)</mo> </mrow> <annotation>$$ left(n=114,000right) $$</annotation> </semantics></math> from Sardinia, Italy, containing text content, topic distributions, and emotional features. Results show significant improvements in both clustering quality (average silhouette score increased) and interpretability compared to traditional single-dimension approaches. From an empirical perspective, the synthetic data validation demonstrates robust performance with topic strength in the range <span></span><math> <semantics> <mrow> <mo>[</mo> <mn>0</mn> <mo>.</mo> <mn>4</mn> <mo>,</mo> <mn>1</mn> <mo>.</mo> <mn>0</mn> <mo>]</mo> </mrow> <annotation>$$ left[0.4,1.0right] $$</annotation> </semantics></math> and emotion strength in <span></span><math> <semantics> <mrow> <mo>[</mo> <mn>0</mn> <mo>.</mo> <mn>2</mn> <mo>,</mo> <mn>1</mn> <mo>.</mo> <mn>0</mn> <mo>]</mo> </mrow> <annotation>$$ left[0.2,1.0right] $$</annotation> </semantics></math>, achieving mean Adjusted Rand Index scores of 0.44. The application to real-world data identifies five distinct clusters through PROCSIMA (PRObabilistic Clustering SIMilarity A
{"title":"Topic-Sentiment Hybrid Networks for Explainable Document Clustering: A Probabilistic Multi-Dimensional Similarity Analysis","authors":"Marco Ortu","doi":"10.1002/asmb.70024","DOIUrl":"https://doi.org/10.1002/asmb.70024","url":null,"abstract":"<p>This study introduces a statistical methodology for document clustering that integrates multiple dimensions of textual similarity through network topology analysis. The proposed methodology, which we call Multi-dimensional Similarity Network Analysis (MSNA), extends traditional document-clustering approaches by combining semantic embeddings, topic probability distributions, and emotional probability distribution into a unified similarity measure. We formalize this through a weighted combination of Jensen-Shannon divergences across different probability spaces, creating a comprehensive similarity network. The clustering is achieved through a community detection algorithm that optimizes a multi-objective modularity function, accounting for the different similarity dimensions. We prove the statistical consistency of our approach and derive bounds for the clustering performance under mild regularity conditions. The methodology is validated on a large-scale data set of Airbnb reviews <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mo>(</mo>\u0000 <mi>n</mi>\u0000 <mo>=</mo>\u0000 <mn>114</mn>\u0000 <mo>,</mo>\u0000 <mn>000</mn>\u0000 <mo>)</mo>\u0000 </mrow>\u0000 <annotation>$$ left(n=114,000right) $$</annotation>\u0000 </semantics></math> from Sardinia, Italy, containing text content, topic distributions, and emotional features. Results show significant improvements in both clustering quality (average silhouette score increased) and interpretability compared to traditional single-dimension approaches. From an empirical perspective, the synthetic data validation demonstrates robust performance with topic strength in the range <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mo>[</mo>\u0000 <mn>0</mn>\u0000 <mo>.</mo>\u0000 <mn>4</mn>\u0000 <mo>,</mo>\u0000 <mn>1</mn>\u0000 <mo>.</mo>\u0000 <mn>0</mn>\u0000 <mo>]</mo>\u0000 </mrow>\u0000 <annotation>$$ left[0.4,1.0right] $$</annotation>\u0000 </semantics></math> and emotion strength in <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mo>[</mo>\u0000 <mn>0</mn>\u0000 <mo>.</mo>\u0000 <mn>2</mn>\u0000 <mo>,</mo>\u0000 <mn>1</mn>\u0000 <mo>.</mo>\u0000 <mn>0</mn>\u0000 <mo>]</mo>\u0000 </mrow>\u0000 <annotation>$$ left[0.2,1.0right] $$</annotation>\u0000 </semantics></math>, achieving mean Adjusted Rand Index scores of 0.44. The application to real-world data identifies five distinct clusters through PROCSIMA (PRObabilistic Clustering SIMilarity A","PeriodicalId":55495,"journal":{"name":"Applied Stochastic Models in Business and Industry","volume":"41 4","pages":""},"PeriodicalIF":1.3,"publicationDate":"2025-06-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/asmb.70024","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144339419","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Industrial processes generate a massive amount of monitoring data that can be exploited to uncover hidden time losses in the system. This can be used to enhance the accuracy of maintenance policies and increase the effectiveness of the equipment. In this work, we propose a method for one-step probabilistic multivariate forecasting of time variables involved in a production process. The method is based on an Input-Output Hidden Markov Model (IO-HMM), in which the parameters of interest are the state transition probabilities and the parameters of the observations' joint density. The ultimate goal of the method is to predict operational process times in the near future, which enables the identification of hidden losses and the location of improvement areas in the process. The input stream in the IO-HMM model includes past values of the response variables and other process features, such as calendar variables, that can have an impact on the model's parameters. The discrete part of the IO-HMM models the operational mode of the process. The state transition probabilities are supposed to change over time and are updated using Bayesian principles. The continuous part of the IO-HMM models the joint density of the response variables. The estimate of the continuous model parameters is recursively computed through an adaptive algorithm that also admits a Bayesian interpretation. The adaptive algorithm allows for efficient updating of the current parameter estimates as soon as new information is available. We evaluate the method's performance using a real data set obtained from a company in a particular sector, and the results are compared with a collection of benchmark models.
{"title":"An Adaptive Learning Approach to Multivariate Time Forecasting in Industrial Processes","authors":"Fernando Miguelez, Josu Doncel, M. D. Ugarte","doi":"10.1002/asmb.70016","DOIUrl":"https://doi.org/10.1002/asmb.70016","url":null,"abstract":"<p>Industrial processes generate a massive amount of monitoring data that can be exploited to uncover hidden time losses in the system. This can be used to enhance the accuracy of maintenance policies and increase the effectiveness of the equipment. In this work, we propose a method for one-step probabilistic multivariate forecasting of time variables involved in a production process. The method is based on an Input-Output Hidden Markov Model (IO-HMM), in which the parameters of interest are the state transition probabilities and the parameters of the observations' joint density. The ultimate goal of the method is to predict operational process times in the near future, which enables the identification of hidden losses and the location of improvement areas in the process. The input stream in the IO-HMM model includes past values of the response variables and other process features, such as calendar variables, that can have an impact on the model's parameters. The discrete part of the IO-HMM models the operational mode of the process. The state transition probabilities are supposed to change over time and are updated using Bayesian principles. The continuous part of the IO-HMM models the joint density of the response variables. The estimate of the continuous model parameters is recursively computed through an adaptive algorithm that also admits a Bayesian interpretation. The adaptive algorithm allows for efficient updating of the current parameter estimates as soon as new information is available. We evaluate the method's performance using a real data set obtained from a company in a particular sector, and the results are compared with a collection of benchmark models.</p>","PeriodicalId":55495,"journal":{"name":"Applied Stochastic Models in Business and Industry","volume":"41 3","pages":""},"PeriodicalIF":1.3,"publicationDate":"2025-06-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/asmb.70016","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144244145","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号