Explore chapters and articles related to this topic
Data learning and expert judgment in a bayesian belief network for offshore decommissioning risk assessment
Published in Stein Haugen, Anne Barros, Coen van Gulijk, Trond Kongsvik, Jan Erik Vinnem, Safety and Reliability – Safe Societies in a Changing World, 2018
M.L. Fam, X.H. He, P. Hilber, L.S. Ong, D. Konovessis, H. K. Tan
Offshore safety data is based mainly on response and explanatory variables of non-metric, discrete forms, such as operation sections on a platform (Drilling and/or Production area) or initiating events of accidents (Falling Load, Leak etc). A common response variable would refer to consequence status differing in levels of severity, and in this case, the order of the status is important, such as ‘Fatality’ having more severe considerations than a ‘Near Miss’. A statistical model suitable for offshore safety data should be required to be able to handle multivariate data or non-metric format, and hence the classical regression models are unsuitable. The proposed statistical model able to handle ordered, categorical data is a log-linear model that falls under the classification of a Generalised Linear Model (GLM) (Agresti, 2002, p. 125). A GLM extend ordinary regression models to encompass non-normal response distributions and modelling functions of the mean and has three components (similar to classical regression models) consisting of a random component (response variable), a systematic component (explanatory variables) and a link function that transforms the mean to the natural parameter. The categorical data is arranged in a table form with its frequency information, and the log-linear modelling involves fitting models to the cell count in the cross-tabulation of categorical variables to derive the association and interaction patterns among variables.
An interpretable model for bridge scour risk assessment using explainable artificial intelligence and engineers’ expertise
Published in Structure and Infrastructure Engineering, 2023
Tianyu Wang, Philippe Reiffsteck, Christophe Chevalier, Chi-Wei Chen, Franziska Schmidt
In this study, multi expression programming (MEP) (Oltean & Dumitrescu, 2002) and generalized linear model (GLM) are used to build two surrogate models. MEP, as a branch of genetic programming (GP) inspired by Charles Darwin’s theory of natural selection, is a type of linear-based GP for optimization. Multiple solutions (programs) are encoded in the same chromosome in MEP. It starts by generating random population of computer programs and the best solution can be generated from the chromosome by iterating the selection, crossover, and mutation process until the termination condition is satisfied. GLM is a linear model that generalizes variables in both numerical and categorical forms via a link function. Compared with a linear regression model in which both variables and outcomes are assumed to follow a Gaussian distribution, GLM allows variables and outcomes to follow the exponential family of distributions (e.g., Poisson distribution, binomial distribution). The link function relates the linear predictor and the mean of the distribution function. For example, the binomial distributed data may use the logistic link function.
Transparent-AI Blueprint: Developing a Conceptual Tool to Support the Design of Transparent AI Agents
Published in International Journal of Human–Computer Interaction, 2022
Zhibin Zhou, Zhuoshu Li, Yuyang Zhang, Lingyun Sun
The entire architecture and working process of these algorithms could be described and understood by humans. The algorithms below offer intrinsic interpretability. (1) The linear regression model predicts the target as a weighted sum of input features, which is easy to interpret. (2) The generalized linear model (GLM) extends the linear regression model. The GLM could model all types of outcomes including category and count. (3) The generalized additive model (GAM) can not only retain linear models’ decent built-in interpretability (Wood, 2006) but also be used to learn nonlinear relationships. (4) The decision rule is an IF-THEN statement that consists of a condition (IF) and prediction (THEN). For rules derived from intelligible features, the IF-THEN structure has a semantic similarity to natural language and human thinking. (5) The decision tree is a tree-like structure in which each non-leaf node represents a feature (attribute), each link (branch) represents a decision (rule), and each leaf represents an outcome (categorical or continues value), which is similar to human thinking.
The impact of sex on motor vehicle crash injury outcomes
Published in Journal of Transportation Safety & Security, 2022
Alyssa Ryan, Francis Tainter, Cole Fitzpatrick, Jennifer Gazzillo, Robin Riessman, Michael Knodler
Generalized linear modeling (GLM) is a framework for statistical analysis that applies regression for special cases. A GLM is given by Equation 1. for where yi is the data vector, Xi are the predictor variables with coefficients β to form a linear predictor and is the error term (Gelman & Hill, 2006). In this research, the response variable yi had the outcomes of y = 1 or y = 0 with respective probability of P and This logistic regression can be written as presented in Equation 2. where β0 is the intercept or constant value, βn is the regression coefficient of Xn, and Xn is the predictor variable. Unlike linear regression which predicts the value that yi takes, logistic regression predicts the probability of yi = 1 for given explanatory variable values. These can be theoretically written as a binary logistic regression, as presented in Equation 3.