China
Africa
Explore chapters and articles related to this topic
Clinical Workflows Supported by Patient Care Device Data
Published in John R. Zaleski, Clinical Surveillance, 2020
Yet, data can tell a story. Time-series data, when taken into account with other contextual information surrounding a patient, can provide an indication of trends that would not be visible when viewed as discrete pieces of information, such as standalone alarm signals. Comparison among data elements over time provides not only the absolute value (e.g., the behavior of interval variables) but also the relationship with respect to the earlier time (e.g., behavior of ratio variables).
Fundamentals
Published in Arvind Kumar Bansal, Javed Iqbal Khan, S. Kaisar Alam, Introduction to Computational Health Informatics, 2019
Arvind Kumar Bansal, Javed Iqbal Khan, S. Kaisar Alam
Time-series data is a periodic discrete sampling of continuous data. There are multiple applications of time-series data such as: 1) studying disease progression, 2) long-term treatment of chronic diseases, 3) recovery after a surgery, 4) managing age-specific diseases and 5) efficacy analysis of new drugs. The nature of time-series data includes big data, multidimensionality of data and frequent updates.
Is there convergence in the health expenditures of the EU Member States?
Published in Elias Mossialos, Julian Le Grand, Health Care and Cost Containment in the European Union, 2019
The alternative approach proposed here is to use a convergence test based in structural time series analysis. This type of analysis is very useful to find the 'underlying trend' of the time series data. The trend can be 'smoothed' by separating its 'true' components (the level and the slope) from its irregular component. This provides a much better picture of where the series 'is going' and how it is 'getting there', and makes any test based on the smoothed (or estimated) trends much more robust.
Beta-negative binomial nonlinear spatio-temporal random effects modeling of COVID-19 case counts in Japan
Published in Journal of Applied Statistics, 2023
The prediction models were evaluated by repeated one-step-ahead prediction, i.e. whether the model trained on past time-series data was able to predict the number of case counts observed the next day. The one-step-ahead predictions were made by sequentially repeating the most-recent past 90 days. The models are all probabilistic models, and the evaluation was done in terms of predictive distributions rather than point prediction. Supplementary Figures S8–S11 give the results of 90 one-step-ahead predictions for the 47 prefectures. Five curves corresponding to 99.9%, 99%, 50%, 1%, and 0.1% quantiles of the prediction models are depicted together with the number of observed cases to be predicted. Perhaps due to its heavy-tailed nature, the beta-negative binomial distribution produced slightly wider prediction intervals than the negative binomial distribution in both the univariate and Paul-Held models. Days where the number of cases exceeded the 99.995% quantile are emphasized; many cases of models using the negative binomial distribution exceeded this threshold. To facilitate visualizing this in greater detail, Figure 2 shows enlarged panels for two prefectures in which extreme values were observed, Nara and Kagawa, chosen from Supplementary Figures S8–S11. Models with the beta-negative binomial distribution properly covered the extreme observations.
Analysis and forecasting of air quality index based on satellite data
Published in Inhalation Toxicology, 2023
Tinku Singh, Nikhil Sharma, Manish Kumar
Holt-Winters extended the Holt model in the 1960s and are now widely utilized in signal processing techniques and forecasting. Holt-Winters is a statistical time-series data behavioral model. There are three components of a time series represented by this metric: average (typical) value, trend (slope) through time, and seasonality (circular repetition pattern). The additive method for the forecasting model is: h is the integer component of the Lt, trend dt, and seasonal component ft, each with its smoothing coefficients α, β.
A method for detection of Mode-Mixing problem
Published in Journal of Applied Statistics, 2021
Atacan Erdiş, M. Akif Bakir, Muhammed I. Jaiteh
A time series is defined as the collection of random variables indexed in the order in which they are obtained over time. Time series analysis is mainly performed to predict and understand the mechanism of the data generating process. To fully understand the mechanism of the data generating process is possible only by determining the characteristic structure of the time series. Traditionally, methods of time series data analysis are mostly based on the assumptions of linearity and stationarity. Analysis methods such as classical and exponential smoothing methods, Box–Jenkins approach and Fourier Transform are mainly the best-known analysis methods based on the assumption of linearity and stationary [3]. Thus, there is familiarity with methods that are designed to analyse linear but non-stationary data (wavelet analysis and Wagner–Ville distribution), as well as non-linear time series analysis methods built for non-linear but stationary data. These methods result to spurious results when applied on non-linear and non-stationary data [12].