UNIGE document Scientific Article
previous document  unige:29127  next document
add to browser collection
Title

Empirical mode decomposition and k-nearest embedding vectors for timely analyses of antibiotic resistance trends

Authors
Published in PLOS ONE. 2013, vol. 8, no. 4, p. e61180
Abstract Background: Antibiotic resistance is a major worldwide public health concern. In clinical settings, timely antibiotic resistance information is key for care providers as it allows appropriate targeted treatment or improved empirical treatment when the specific results of the patient are not yet available. Objective: To improve antibiotic resistance trend analysis algorithms by building a novel, fully data-driven forecasting method from the combination of trend extraction and machine learning models for enhanced biosurveillance systems. Methods: We investigate a robust model for extraction and forecasting of antibiotic resistance trends using a decade of microbiology data. Our method consists of breaking down the resistance time series into independent oscillatory components via the empirical mode decomposition technique. The resulting waveforms describing intrinsic resistance trends serve as the input for the forecasting algorithm. The algorithm applies the delay coordinate embedding theorem together with the k-nearest neighbor framework to project mappings from past events into the future dimension and estimate the resistance levels. Results: The algorithms that decompose the resistance time series and filter out high frequency components showed statistically significant performance improvements in comparison with a benchmark random walk model. We present further qualitative use-cases of antibiotic resistance trend extraction, where empirical mode decomposition was applied to highlight the specificities of the resistance trends. Conclusion: The decomposition of the raw signal was found not only to yield valuable insight into the resistance evolution, but also to produce novel models of resistance forecasters with boosted prediction performance, which could be utilized as a complementary method in the analysis of antibiotic resistance trends.
Identifiers
PMID: 23637796
Full text
Article (Published version) (828 Kb) - public document Free access
Structures
Research group Interfaces Homme-machine en milieu clinique (610)
Project FP7: DebugIT
Citation
(ISO format)
TEODORO, Douglas Henrique, LOVIS, Christian. Empirical mode decomposition and k-nearest embedding vectors for timely analyses of antibiotic resistance trends. In: PLOS ONE, 2013, vol. 8, n° 4, p. e61180. https://archive-ouverte.unige.ch/unige:29127

229 hits

100 downloads

Update

Deposited on : 2013-08-09

Export document
Format :
Citation style :