Article
Mathematics, Interdisciplinary Applications
Gerhard Tutz
Summary: The study introduces an improved method for ordinal trees that avoid the artificial assignment of scores and adopts the construction principle of binary models, combining trees and parametric models for prediction. The potential performance issues of random forests are also discussed, with proposals for ensemble models to achieve better predictive performance.
JOURNAL OF CLASSIFICATION
(2022)
Article
Biochemistry & Molecular Biology
Sk Abdul Amin, Nilanjan Adhikari, Tarun Jha
Summary: In this study, a diverse set of compounds were analyzed using recursive partitioning (RP) analysis to develop decision trees for discriminating HDAC8 inhibitors from non-inhibitors. Understanding essential structural and physicochemical parameters is crucial for designing potential and selective HDAC8 inhibitors, and the results validate previous findings from Bayesian modeling. This comparative learning will enhance drug discovery efforts related to HDAC8 inhibitors.
JOURNAL OF BIOMOLECULAR STRUCTURE & DYNAMICS
(2021)
Article
Computer Science, Theory & Methods
Christophe Dutang, Quentin Guibert
Summary: This paper proposes a split point procedure based on explicit likelihood to speed up the search for the best split point in CART. Through simulation and benchmarking on empirical datasets, GLM trees are shown to have good performance in certain situations. The approach is extended to multiway split trees and log-transformed distributions. A numerical comparison of GLM forests against other random forest-type approaches is also provided.
STATISTICS AND COMPUTING
(2022)
Article
Statistics & Probability
Naoki Awaya, Li Ma
Summary: The Polya tree (PT) process is a versatile Bayesian nonparametric model that has been widely used in inference problems. Recent developments have shown that the performance of PT models can be improved by adapting the partition tree to the underlying distributions and incorporating latent state variables. However, there are still important limitations, including sensitivity to the choice of the partition tree and lack of scalability with respect to dimensionality.
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
(2022)
Article
Ergonomics
Ali Khodadadi, Ioannis Tsapakis, Mohammadali Shirazi, Subasish Das, Dominique Lord
Summary: This study proposed an Empirical Bayesian method based on the Negative Binomial-Lindley model for estimating the expected crash frequency. The results showed that this method can estimate the expected crashes with comparable precision to the Full Bayesian method, but with lower computational cost. It can be applied to other safety-related tasks.
ACCIDENT ANALYSIS AND PREVENTION
(2022)
Article
Statistics & Probability
Sonja Isberg, William J. Welch
Summary: This study discusses multiple methods for addressing the computational complexity of computer models in large designs, and proposes a new method called "adaptive design and analysis via partitioning trees (ADAPT)". The proposed method partitions the input space in regions of high variability to obtain a higher density of points for accurate prediction.
JOURNAL OF COMPUTATIONAL AND GRAPHICAL STATISTICS
(2022)
Article
Automation & Control Systems
Litao Zheng, Feng Yang, Lihong Shi
Summary: This paper proposes a fault-tolerant distributed Bayesian filter for multi-sensor state estimation using a peer-to-peer sensor network with incoherent local estimates problems. The proposed approach uses a Gaussian mixture to represent the fusion result, effectively reducing negative impact. The resulting filter performs Bayesian recursion via Gaussian mixture and utilizes a novel arithmetic average fusion for heterogeneous sensor networks.
Article
Construction & Building Technology
Giosue Boscato, Marco Civera, Luca Zanotti Fragonara
Summary: The paper proposes a methodology for detecting and localizing damages in composite pultruded members, particularly focusing on thin-walled pultruded members. The method is applied to numerical and experimental data, analyzing both modal shapes and the influence of damage on the performance of Glass Fiber Reinforced Polymer (GFRP) members. The reliability of the proposed semiparametric statistical method is demonstrated through numerical investigation and comparison with experimental results on cracked beams and frame structures.
STRUCTURAL CONTROL & HEALTH MONITORING
(2021)
Article
Ecology
Joseph M. Eisaguirre, Travis L. Booms, Christopher P. Barger, Stephen B. Lewis, Greg A. Breed
Summary: This study explores the differential habitat selection and space use between floaters and territorial golden eagles based on satellite telemetry data. The results reveal that floaters have more expansive space use patterns and larger home ranges compared to territorial eagles, and they partition space with territorial individuals through differential habitat and resource selection.
ECOLOGICAL APPLICATIONS
(2022)
Article
Linguistics
Lukas Soenning, Jason Grafmiller
Summary: Classification trees and random forests are attractive methods for corpus data analysis. However, their typical reporting style lacks sufficient information on the relationship between predictors and outcomes. This paper introduces predictive margins as an interpretative approach to ensemble techniques like random forests, providing adjusted predictions and allowing for nonlinear associations and interactions. It outlines the general strategy and addresses methodological issues, using English genitive alternation data as an example and providing an R package for implementation.
CORPUS LINGUISTICS AND LINGUISTIC THEORY
(2023)
Article
Statistics & Probability
Christian Bartels, Johanna Mielke, Ekkehard Glimm
Summary: This study proposes a frequentist testing procedure that allows adjusting the decision rules and increasing power by selecting a prior distribution. However, it comes with the risk of losing power if the data generating distribution or the observed data are incompatible with the prior distribution. The approach is illustrated using a simple binomial experiment and the potential beyond the example is discussed. It is worth noting that the testing procedure is constructed using Bayesian posterior probability distribution.
COMMUNICATIONS IN STATISTICS-THEORY AND METHODS
(2022)
Article
Computer Science, Artificial Intelligence
Wenhan Fu, Chen-Fu Chien, Lizhen Tang
Summary: Probe cards are essential test interfaces for integrated circuit testing, but the diagnosing and troubleshooting process can be complex and time-consuming. This study aims to develop a Bayesian network using data-driven solutions and potential rules derived from domain knowledge to enhance data integrity and improve troubleshooting efficiency.
JOURNAL OF INTELLIGENT MANUFACTURING
(2022)
Article
Multidisciplinary Sciences
Warisa Thangjai, Sa-Aat Niwitpong, Suparat Niwitpong
Summary: Weighted percentiles are used to investigate the overall trend of rainfall in Thailand, and confidence intervals for common percentiles of delta-lognormal distributions are constructed. Comparisons of coverage probabilities and average lengths show that one Bayesian approach performs better than others.
Article
Computer Science, Artificial Intelligence
S. Jozova, E. Uglickich, I. Nagy, R. Likhonina
Summary: The paper presents an algorithm for modeling discrete questionnaire data with reduced dimension. The algorithm reduces the dimension of the discrete model by constructing local models based on independent binomial mixtures estimated using recursive Bayesian algorithms and the naive Bayes technique. The algorithm allows for modeling high dimensional questionnaire data with a large number of explanatory variables and their possible realizations. The algorithm is applied to the analysis of traffic accident questionnaires, where it is used for classifying accident circumstances and predicting the severity of traffic accidents using current discrete data. The effectiveness of the obtained model is demonstrated through testing on real data and comparison with theoretical counterparts.
NEURAL NETWORK WORLD
(2022)
Article
Computer Science, Theory & Methods
Valentin De Bortoli, Alain Durmus, Marcelo Pereyra, Ana F. Vidal
Summary: This paper proposes a method using unadjusted Langevin algorithms to construct stochastic approximation, addressing the difficulties of using high-dimensional Markov chain Monte Carlo algorithms in large problems. This approach leads to a highly efficient stochastic optimization method with favorable convergence properties that can be quantified explicitly and easily checked.
STATISTICS AND COMPUTING
(2021)