scholarly journals Regional regression models of percentile flows for the contiguous United States: Expert versus data-driven independent variable selection

2018 ◽  
Vol 17 ◽  
pp. 64-82 ◽  
Author(s):  
Geoffrey Fouad ◽  
André Skupin ◽  
Christina L. Tague
Keyword(s):  
United States ◽  
Variable Selection ◽  
Regression Models ◽  
Data Driven ◽  
Versus Data ◽  
Regional Regression ◽  
Independent Variable

scholarly journals Regional regression models of percentile flows for the contiguous US: Expert versus data-driven independent variable selection

2016 ◽  
Author(s):  
Geoffrey Fouad ◽  
André Skupin ◽  
Christina L. Tague
Keyword(s):  
Regression Model ◽  
Regression Models ◽  
Predictive Performance ◽  
Data Driven ◽  
Mean Annual Precipitation ◽  
Expert Assessment ◽  
Independent Variables ◽  
Regional Regression ◽  
Data Driven Approach ◽  
Small Set

Abstract. Percentile flows are statistics derived from the flow duration curve (FDC) that describe the flow equaled or exceeded for a given percent of time. These statistics provide important information for managing rivers, but are often unavailable since most basins are ungauged. A common approach for predicting percentile flows is to deploy regional regression models based on gauged percentile flows and related independent variables derived from physical and climatic data. The first step of this process identifies groups of basins through a cluster analysis of the independent variables, followed by the development of a regression model for each group. This entire process hinges on the independent variables selected to summarize the physical and climatic state of basins. Distributed physical and climatic datasets now exist for the contiguous United States (US). However, it remains unclear how to best represent these data for the development of regional regression models. The study presented here developed regional regression models for the contiguous US, and evaluated the effect of different approaches for selecting the initial set of independent variables on the predictive performance of the regional regression models. An expert assessment of the dominant controls on the FDC was used to identify a small set of independent variables likely related to percentile flows. A data-driven approach was also applied to evaluate two larger sets of variables that consist of either (1) the averages of data for each basin or (2) both the averages and statistical distribution of basin data distributed in space and time. The small set of variables from the expert assessment of the FDC and two larger sets of variables for the data-driven approach were each applied for a regional regression procedure. Differences in predictive performance were evaluated using 184 validation basins withheld from regression model development. The small set of independent variables selected through expert assessment produced similar, if not better, performance than the two larger sets of variables. A parsimonious set of variables only consisted of mean annual precipitation, potential evapotranspiration, and baseflow index. Additional variables in the two larger sets of variables added little to no predictive information. Regional regression models based on the parsimonious set of variables were developed using 734 calibration basins, and were converted into a tool for predicting 13 percentile flows in the contiguous US. Supplementary Material for this paper includes an R graphical user interface for predicting the percentile flows of basins within the range of conditions used to calibrate the regression models. The equations and performance statistics of the models are also supplied in tabular form.

scholarly journals Sparse Regression in Cancer Genomics: Comparing Variable Selection and Predictions in Real World Data

2021 ◽  
Vol 20 ◽  
pp. 117693512110562
Author(s):  
Robert J O’Shea ◽  
Sophia Tsoka ◽  
Gary JR Cook ◽  
Vicky Goh
Keyword(s):  
Variable Selection ◽  
Test Performance ◽  
Gold Standard ◽  
Regression Models ◽  
Cross Validation ◽  
Cancer Genomics ◽  
Genomic Data ◽  
Data Driven ◽  
Real World Data ◽  
Standard Models

Background: Evaluation of gene interaction models in cancer genomics is challenging, as the true distribution is uncertain. Previous analyses have benchmarked models using synthetic data or databases of experimentally verified interactions – approaches which are susceptible to misrepresentation and incompleteness, respectively. The objectives of this analysis are to (1) provide a real-world data-driven approach for comparing performance of genomic model inference algorithms, (2) compare the performance of LASSO, elastic net, best-subset selection, [Formula: see text] penalisation and [Formula: see text] penalisation in real genomic data and (3) compare algorithmic preselection according to performance in our benchmark datasets to algorithmic selection by internal cross-validation. Methods: Five large [Formula: see text] genomic datasets were extracted from Gene Expression Omnibus. ‘Gold-standard’ regression models were trained on subspaces of these datasets ([Formula: see text], [Formula: see text]). Penalised regression models were trained on small samples from these subspaces ([Formula: see text]) and validated against the gold-standard models. Variable selection performance and out-of-sample prediction were assessed. Penalty ‘preselection’ according to test performance in the other 4 datasets was compared to selection internal cross-validation error minimisation. Results: [Formula: see text]-penalisation achieved the highest cosine similarity between estimated coefficients and those of gold-standard models. [Formula: see text]-penalised models explained the greatest proportion of variance in test responses, though performance was unreliable in low signal:noise conditions. [Formula: see text] also attained the highest overall median variable selection F1 score. Penalty preselection significantly outperformed selection by internal cross-validation in each of 3 examined metrics. Conclusions: This analysis explores a novel approach for comparisons of model selection approaches in real genomic data from 5 cancers. Our benchmarking datasets have been made publicly available for use in future research. Our findings support the use of [Formula: see text] penalisation for structural selection and [Formula: see text] penalisation for coefficient recovery in genomic data. Evaluation of learning algorithms according to observed test performance in external genomic datasets yields valuable insights into actual test performance, providing a data-driven complement to internal cross-validation in genomic regression tasks.

scholarly journals Research on a novel data-driven aging estimation method for battery systems in real-world electric vehicles

2021 ◽  
Vol 13 (7) ◽  
pp. 168781402110277
Author(s):  
Yankai Hou ◽  
Zhaosheng Zhang ◽  
Peng Liu ◽  
Chunbao Song ◽  
Zhenpo Wang
Keyword(s):  
Electric Vehicles ◽  
Real World ◽  
Regression Models ◽  
Estimation Method ◽  
Recursive Least Squares ◽  
Data Driven ◽  
Accurate Estimation ◽  
Support Vector ◽  
Battery Degradation ◽  
Operational Data

Accurate estimation of the degree of battery aging is essential to ensure safe operation of electric vehicles. In this paper, using real-world vehicles and their operational data, a battery aging estimation method is proposed based on a dual-polarization equivalent circuit (DPEC) model and multiple data-driven models. The DPEC model and the forgetting factor recursive least-squares method are used to determine the battery system’s ohmic internal resistance, with outliers being filtered using boxplots. Furthermore, eight common data-driven models are used to describe the relationship between battery degradation and the factors influencing this degradation, and these models are analyzed and compared in terms of both estimation accuracy and computational requirements. The results show that the gradient descent tree regression, XGBoost regression, and light GBM regression models are more accurate than the other methods, with root mean square errors of less than 6.9 mΩ. The AdaBoost and random forest regression models are regarded as alternative groups because of their relative instability. The linear regression, support vector machine regression, and k-nearest neighbor regression models are not recommended because of poor accuracy or excessively high computational requirements. This work can serve as a reference for subsequent battery degradation studies based on real-time operational data.

scholarly journals Correction to: MicroBVS: Dirichlet-tree multinomial regression models with Bayesian variable selection - an R package

2020 ◽  
Vol 21 (1) ◽  
Author(s):  
Matthew D. Koslovsky ◽  
Marina Vannucci
Keyword(s):  
Variable Selection ◽  
Regression Models ◽  
R Package ◽  
Bayesian Variable Selection ◽  
Multinomial Regression

An amendment to this paper has been published and can be accessed via the original article.

Sign in / Sign up

Export Citation Format

Share Document