China
Africa
Explore chapters and articles related to this topic
The Importance of Intelligent HIT Vendors
Published in Tom Lawry, AI in Health, 2020
Monitoring and managing all predictive models that have been deployed. An area taking on greater importance as predictive capabilities become more mainstream is the systemic monitoring of all algorithms to ensure that they are used in keeping with an organization’s clinical, regulatory, and ethical standards. How predictive models perform varies based on many factors such as data itself or the environment in which predictive models operate. For example, the use of a deployed predictive model for clinical decision support might be affected by the addition of new hospitals whose data is added as part of a merger or acquisition. Another example is where a predictive value may demonstrate overall effectiveness in improving performance but have wide variance in effectiveness based on variables such as age, gender, and race. Systems that automatically monitor and safeguard such inconsistencies are critical to the success of AI being deployed at scale.
Predictive modelling
Published in Catherine Dawson, A–Z of Digital Research Methods, 2019
Methods used to generate predictive models require two types of data: predictors (or inputs) and outcomes or outputs (the behaviour that is to be predicted). An appropriate statistical or mathematical technique is applied to the data to determine relationships, which are captured in the resulting model. This model can be applied to situations where the predictors are known, but where outputs are unknown. There are two broad classes of predictive models: parametric models that have a finite number of parameters, or known inputs, and nonparametric models that have a (potentially) infinite number of parameters, or many different inputs (parameters are flexible and not fixed in advance). It is also possible to have semi-parametric models that contain components from both of these models.
Radiogenomics
Published in Jun Deng, Lei Xing, Big Data in Radiation Oncology, 2019
Barry S. Rosenstein, Gaurav Pandey, Corey W. Speers, Jung Hun Oh, Catharine M.L. West, Charles S. Mayo
Supervised learning includes machine learning analyses that utilize not only the features of data points, but also additional information (supervision) available about them. Examples of such information include class labels of data points (e.g., diseased or healthy), functional annotations of genes and disease phenotypes (e.g., survival rates) of patients, among many other possibilities. A common analysis of this type is predictive modeling, which includes analyses like classification and regression. In a typical workflow used for predictive modeling, a learning algorithm is applied to a training data set, where class labels are available for the data objects described in terms of attributes or features, to induce a predictive model. This model captures the relationship between feature values and class labels in a form determined by the choice of the learning algorithm. This model is then applied to unlabeled data objects, possibly constituting one or more test sets, to deduce their class labels. Using methodologies such as cross-validation or training-holdout splits, and standardized measures like area under the receiver operating characteristic (ROC) curve (AUC), precision-recall-F-measure, and mean squared error, the performance of predictive models can be objectively evaluated. For details of the concepts above, we refer the reader to standard textbooks on the topic (Tan et al. 2006; Kuhn and Johnson 2013).
Predicting additive manufacturing defects with robust feature selection for imbalanced data
Published in IISE Transactions, 2023
Ethan Houser, Sara Shashaani, Ola Harrysson, Yongseok Jeon
The main goal of SOEN is to determine which set of the p features to use when constructing a predictive model to minimize its OOB error. Not every feature is a statistically significant indicator of defects. However, manually checking each unique combination of features would be a cumbersome task. Opting to build a predictive model with the entire list of features can lead to overfitting, increased computational complexity for constructing the model, and a significant reduction in the predictive performance (Shashaani and Vahdat, 2022). We use a Genetic Algorithm (GA) as the solver appropriate for the binary search space that generates sets of solutions iteratively and ranks them using the estimated objective function The solutions of SOEN are iteratively refined with a GA. We refer to these iterative solutions as for iteration k. Figure 7 provides an illustration of this situation.
Quantitative prediction of fracture toughness (K Ic ) of polymer by fractography using deep neural networks
Published in Science and Technology of Advanced Materials: Methods, 2022
Y. Mototake, K. Ito, M. Demura
To construct such a regression model, the DNN feature , the kernel function type , and the kernel function hyperparameters must be properly selected. In this study, these parameters are selected on the basis of a Bayesian inference framework [23]. For the selection of and , the Bayesian model selection framework is employed. On the basis of an indicator called Bayesian free energy [24], the Bayesian model selection framework selects the model that considers discrete-valued hyperparameters such as or (see Appendix). In general, when using machine learning to build predictive models, it is necessary to prevent over-fitting, where the model over-fits the training data and loses predictive performance. Using Bayesian free energy as an indicator for model selection, it is possible to select a model in which over-fitting does not occur. , that is, the continuous-valued hyperparameter, can also be estimated on the basis of Bayesian free energy (see Appendix).
Explanatory Model Analysis: Explore, Explain, and Examine Predictive Models
Published in Journal of Quality Technology, 2022
Predictive models aim to guess, a.k.a., predict, values of a variable of interest based on other variables. It has been used throughout the entire human history and many statistical models have been developed for prediction during the last century. This book covers methods for exploration of predictive models from both instance level and dataset level. It would be a valuable addition to the Chapman & Hall/CRC’s Data Science Series. Together with other books that have published in the book series, this book provides a unique perspective into applied data science to guide data science practitioners who are interested in exploring, explaining, and examining data in real-world applications with both R and Python. Predictive models constitute an important component in the big picture of machine learning and data science approaches and require standard analytical steps such as model specification, model estimation, and model fitness diagnosis. Most of published books in this field focus on how to use these statistical methods to make predictions for different types of datasets, while lack of tools for model exploration and, in particular, model explanation (obtaining insights from model-based prediction) and model examination (evaluation of model performance and understanding its weakness). In contrast, this book is a novel effort that provides a deep understanding to all the steps with extensive validation and justification methods, leading to a better and faster interpretable data analysis.