Alternative adjustment for seasonality and long-term time-trend in time-series analysis for long-term environmental exposures and disease counts

Abstract Background Time-series analysis with case-only data is a prominent method for the effect of environmental determinants on disease events in environmental epidemiology. In this analysis, adjustment for seasonality and long-term time-trend is crucial to obtain valid findings. When applying this analysis for long-term exposure (e.g., months, years) of which effects are usually studied via survival analysis with individual-level longitudinal data, unlike its application for short-term exposure (e.g., days, weeks), a standard adjustment method for seasonality and long-term time-trend can extremely inflate standard error of coefficient estimates of the effects. Given that individual-level longitudinal data are difficult to construct and often available to limited populations, if this inflation of standard error can be solved, rich case-only data over regions and countries would be very useful to test a variety of research hypotheses considering unique local contexts. Methods We discuss adjustment methods for seasonality and time-trend used in time-series analysis in environmental epidemiology and explain why standard errors can be inflated. We suggest alternative methods to solve this problem. We conduct simulation analyses based on real data for Seoul, South Korea, 2002–2013, and time-series analysis using real data for seven major South Korean cities, 2006–2013 to identify whether the association between long-term exposure and health outcomes can be estimated via time-series analysis with alternative adjustment methods. Results Simulation analyses and real-data analysis confirmed that frequently used adjustment methods such as a spline function of a variable representing time extremely inflate standard errors of estimates for associations between long-term exposure and health outcomes. Instead, alternative methods such as a combination of functions of variables representing time can make sufficient adjustment with efficiency. Conclusions Our findings suggest that time-series analysis with case-only data can be applied for estimating long-term exposure effects. Rich case-only data such as death certificates and hospitalization records combined with repeated measurements of environmental determinants across countries would have high potentials for investigating the effects of long-term exposure on health outcomes allowing for unique contexts of local populations.

Download Full-text

Long-Term Data Traffic Forecasting for Network Dimensioning in LTE with Short Time Series

Electronics ◽

10.3390/electronics10101151 ◽

2021 ◽

Vol 10 (10) ◽

pp. 1151

Author(s):

Carolina Gijón ◽

Matías Toril ◽

Salvador Luna-Ramírez ◽

María Luisa Marí-Altozano ◽

José María Ruiz-Avilés

Keyword(s):

Time Series ◽

Deep Learning ◽

Time Series Analysis ◽

Data Traffic ◽

Medium Term ◽

Short Term ◽

Traffic Forecasting ◽

Series Analysis ◽

Radio Planning

Network dimensioning is a critical task in current mobile networks, as any failure in this process leads to degraded user experience or unnecessary upgrades of network resources. For this purpose, radio planning tools often predict monthly busy-hour data traffic to detect capacity bottlenecks in advance. Supervised Learning (SL) arises as a promising solution to improve predictions obtained with legacy approaches. Previous works have shown that deep learning outperforms classical time series analysis when predicting data traffic in cellular networks in the short term (seconds/minutes) and medium term (hours/days) from long historical data series. However, long-term forecasting (several months horizon) performed in radio planning tools relies on short and noisy time series, thus requiring a separate analysis. In this work, we present the first study comparing SL and time series analysis approaches to predict monthly busy-hour data traffic on a cell basis in a live LTE network. To this end, an extensive dataset is collected, comprising data traffic per cell for a whole country during 30 months. The considered methods include Random Forest, different Neural Networks, Support Vector Regression, Seasonal Auto Regressive Integrated Moving Average and Additive Holt–Winters. Results show that SL models outperform time series approaches, while reducing data storage capacity requirements. More importantly, unlike in short-term and medium-term traffic forecasting, non-deep SL approaches are competitive with deep learning while being more computationally efficient.

Download Full-text