Comparative study of techniques for large-scale feature selection* *This work was suported by a SERC grant GR/E 97549. The first author was also supported by a FPI grant from the Spanish MEC, PF92 73546684

Ferri, Francesc J.; Pudil, Pavel; Hatef, M.; Kittler, Josef

doi:10.1016/b978-0-444-81892-8.50040-7

Cited by 175 publications

(54 citation statements)

References 1 publication

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…We used a two-way cross-validation on the validation set for hyper-parameters tuning to compute these imaging features. A sequential feature selection ( Ferri et al, 1994 ) was then conducted on the hybrid feature pool. Specifically, starting from an empty set, we picked one feature at a time from the remaining feature pool that minimized a validation loss.…”

Section: Methodsmentioning

confidence: 99%

The FELIX Project: Deep Networks To Detect Pancreatic Neoplasms

Xia

Chu

et al. 2022

Preprint

View full text Add to dashboard Cite

Tens of millions of abdominal images are performed with computed tomography (CT) in the U.S. each year but pancreatic cancers are sometimes not initially detected in these images. We here describe a suite of algorithms (named FELIX) that can recognize pancreatic lesions from CT images without human input. Using FELIX, >90% of patients with pancreatic ductal adenocarcinomas were detected at a specificity of >90% in patients without pancreatic disease. FELIX may be able to assist radiologists in identifying pancreatic cancers earlier, when surgery and other treatments offer more hope for long-term survival.

show abstract

Section: Methodsmentioning

confidence: 99%

The FELIX Project: Deep Networks To Detect Pancreatic Neoplasms

Xia

Chu

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

Section: Foldes Et Almentioning

confidence: 99%

“…The additional features are evaluated by adding recursively the quantity which produces the best variation of the R 2 -score with respect to the previous value. This technique, commonly known as Sequential Forward Selection (SFS, Ferri et al, 1994), represents the most powerful tool to investigate medium and R is the score obtained using only the cross-phase as input features. Methods with lower performance (Kernel Ridge Regression and Support Vector Regression) have an improvement with additional features while, as the model accuracy increases, they become less relevant.…”

Section: Foldes Et Almentioning

confidence: 99%

Assessing Machine Learning Techniques for Identifying Field Line Resonance Frequencies From Cross‐Phase Spectra

et al. 2021

View full text Add to dashboard Cite

The near-Earth environment is composed of different plasma populations that cover an energy range of several orders of magnitude. The plasmasphere is the coldest one (∼1 eV) and typically extends from the top of the ionosphere up to 4-6 Earth radii (R E ). Its shape, composition and extension change in response to geomagnetic activity variation and, due to its preponderant contribution to the mass density, plays a crucial role in exciting plasma-waves and driving their interaction with more energetic particles (Liemohn, 2006;Thorne, 2010). During major storms the outer boundary, the plasmapause, also defined as the plasmasphere boundary layer (PBL, Carpenter & Lemaire, 2004), can move earthward as near as 1.5-2 R E , and the recovery to the pre-storm conditions can last for several days.For its central role in the space weather context, a real-time monitoring of the cold plasma in the inner magnetosphere would be an essential task to achieve. A prototype of such system was developed inside the PLASMON FP7 European project (Lichtenberger et al., 2013) during which the European quasi-Meridional Magnetometer Array (EMMA) was established. EMMA provides a unique opportunity to monitor the plasmasphere dynamics in near real-time. It consists of 27 ground-based magnetic stations approximately located along the same magnetic meridian and mapping into magnetic L-shells in the range 1.6-6.1, where L is the McIlwain parameter evaluated using the IGRF. Using Ultra Low Frequency (ULF) signals simultaneously observed by pairs of stations it is possible to infer the radial distribution of the equatorial plasma mass density in the longitudinal sector identified by the network (Del Corpo et al., 2019;Lichtenberger et al., 2013).The core of this monitoring system is the gradient method proposed by Baransky et al. (1985) and further developed by Waters et al. (1991): By performing Fourier cross-spectral analysis of the magnetic signal recorded by two stations slightly separated in latitude, it is possible to estimate the resonance frequency of the field line whose footprint is halfway between the stations. Around the field line resonance (FLR) frequency

show abstract

“…First, a nonlinear regression tree model, which was made with XGBoost framework (extreme gradient boosting) [27], was used to fit weekly new cases and COVID-19 growth rates. Next, sequential backward floating selection was iterated to train the XGBoost [28,29] model to obtain the final model by minimizing the mean squared error. Sequential backward floating selection is a sequential feature selection method based on a greedy search algorithm.…”

Section: Feature Selection and Feature Importancementioning

confidence: 99%

“…As a result, a nonlinear method was needed to assess the contribution of correlated factors on the spread of COVID-19 in China. Machine learning methods are good for solving nonlinear problems, and we used XGBoost [28,30] to create a nonlinear regression tree model for this study. Important factors were selected based on the cross-validation procedure in the XGBoost framework.…”

Section: Multifactor Analysis Based On the Machine Learning Algorithmmentioning

confidence: 99%

Impact of Systematic Factors on the Outbreak Outcomes of the Novel COVID-19 Disease in China: Factor Analysis Study (Preprint)

Cao¹,

Tang²,

Cai³

et al. 2020

Preprint

View full text Add to dashboard Cite

BACKGROUND The novel COVID-19 disease has spread worldwide, resulting in a new pandemic. The Chinese government implemented strong intervention measures in the early stage of the epidemic, including strict travel bans and social distancing policies. Prioritizing the analysis of different contributing factors to outbreak outcomes is important for the precise prevention and control of infectious diseases. We proposed a novel framework for resolving this issue and applied it to data from China. OBJECTIVE This study aimed to systematically identify national-level and city-level contributing factors to the control of COVID-19 in China. METHODS Daily COVID-19 case data and related multidimensional data, including travel-related, medical, socioeconomic, environmental, and influenza-like illness factors, from 343 cities in China were collected. A correlation analysis and interpretable machine learning algorithm were used to evaluate the quantitative contribution of factors to new cases and COVID-19 growth rates during the epidemic period (ie, January 17 to February 29, 2020). RESULTS Many factors correlated with the spread of COVID-19 in China. Travel-related population movement was the main contributing factor for new cases and COVID-19 growth rates in China, and its contributions were as high as 77% and 41%, respectively. There was a clear lag effect for travel-related factors (previous vs current week: new cases, 45% vs 32%; COVID-19 growth rates, 21% vs 20%). Travel from non-Wuhan regions was the single factor with the most significant impact on COVID-19 growth rates (contribution: new cases, 12%; COVID-19 growth rate, 26%), and its contribution could not be ignored. City flow, a measure of outbreak control strength, contributed 16% and 7% to new cases and COVID-19 growth rates, respectively. Socioeconomic factors also played important roles in COVID-19 growth rates in China (contribution, 28%). Other factors, including medical, environmental, and influenza-like illness factors, also contributed to new cases and COVID-19 growth rates in China. Based on our analysis of individual cities, compared to Beijing, population flow from Wuhan and internal flow within Wenzhou were driving factors for increasing the number of new cases in Wenzhou. For Chongqing, the main contributing factor for new cases was population flow from Hubei, beyond Wuhan. The high COVID-19 growth rates in Wenzhou were driven by population-related factors. CONCLUSIONS Many factors contributed to the COVID-19 outbreak outcomes in China. The differential effects of various factors, including specific city-level factors, emphasize the importance of precise, targeted strategies for controlling the COVID-19 outbreak and future infectious disease outbreaks.

show abstract

Comparative study of techniques for large-scale feature selection* *This work was suported by a SERC grant GR/E 97549. The first author was also supported by a FPI grant from the Spanish MEC, PF92 73546684

Cited by 175 publications

References 1 publication

The FELIX Project: Deep Networks To Detect Pancreatic Neoplasms

The FELIX Project: Deep Networks To Detect Pancreatic Neoplasms

Assessing Machine Learning Techniques for Identifying Field Line Resonance Frequencies From Cross‐Phase Spectra

Impact of Systematic Factors on the Outbreak Outcomes of the Novel COVID-19 Disease in China: Factor Analysis Study (Preprint)

Contact Info

Product

Resources

About