Yulia R. Gel

Forecasting influenza in Hong Kong with Google search queries and statistical model fusion

Background The objective of this study is to investigate predictive utility of online social media and web search queries, particularly, Google search data, to forecast new cases of influenza-like-illness (ILI) in general outpatient clinics (GOPC) in Hong Kong. To mitigate the impact of sensitivity to self-excitement (i.e., fickle media interest) and other artifacts of online social media data, in our approach we fuse multiple offline and online data sources. Methods Four individual models: generalized linear model (GLM), least absolute shrinkage and selection operator (LASSO), autoregressive integrated moving average (ARIMA), and deep learning (DL) with Feedforward Neural Networks (FNN) are employed to forecast ILI-GOPC both one week and two weeks in advance. The covariates include Google search queries, meteorological data, and previously recorded offline ILI. To our knowledge, this is the first study that introduces deep learning methodology into surveillance of infectious diseases and investigates its predictive utility. Furthermore, to exploit the strength from each individual forecasting models, we use statistical model fusion, using Bayesian model averaging (BMA), which allows a systematic integration of multiple forecast scenarios. For each model, an adaptive approach is used to capture the recent relationship between ILI and covariates. Results DL with FNN appears to deliver the most competitive predictive performance among the four considered individual models. Combing all four models in a comprehensive BMA framework allows to further improve such predictive evaluation metrics as root mean squared error (RMSE) and mean absolute predictive error (MAPE). Nevertheless, DL with FNN remains the preferred method for predicting locations of influenza peaks. Conclusions The proposed approach can be viewed a feasible alternative to forecast ILI in Hong Kong or other countries where ILI has no constant seasonal trend and influenza data resources are limited. The proposed methodology is easily tractable and computationally efficient.

Multilevel Random Slope Approach and Nonparametric Inference for River Temperature, Under Haphazard Sampling

Environmental scientists face multiple challenges when analyzing unevenly recorded time series with small sample sizes. For example, trends in water temperature may be confounded with time and date of sampling when the latter represent convenience samples and thus introduce bias into regression estimates. We address these concerns using multilevel random slope models and nonparametric bootstrap inference for assessing the statistical significance of the annual trend in river temperature when measurement times and dates are haphazard.

Bootstrap quantification of estimation uncertainties in network degree distributions

We propose a new method of nonparametric bootstrap to quantify estimation uncertainties in functions of network degree distribution in large ultra sparse networks. Both network degree distribution and network order are assumed to be unknown. The key idea is based on adaptation of the “blocking” argument, developed for bootstrapping of time series and re-tiling of spatial data, to random networks. We first sample a set of multiple ego networks of varying orders that form a patch, or a network block analogue, and then resample the data within patches. To select an optimal patch size, we develop a new computationally efficient and data-driven cross-validation algorithm. The proposed fast patchwork bootstrap (FPB) methodology further extends the ideas for a case of network mean degree, to inference on a degree distribution. In addition, the FPB is substantially less computationally expensive, requires less information on a graph, and is free from nuisance parameters. In our simulation study, we show that the new bootstrap method outperforms competing approaches by providing sharper and better-calibrated confidence intervals for functions of a network degree distribution than other available approaches, including the cases of networks in an ultra sparse regime. We illustrate the FPB in application to collaboration networks in statistics and computer science and to Wikipedia networks.

A distribution-free m -out-of- n bootstrap approach to testing symmetry about an unknown median

Testing for symmetry about an unknown median is a ubiquitous problem in mathematical statistics, particularly, for nonparametric rank-based methods, and in a broad range of applied studies, from economics and business to biology, ecology, and medicine. However, the challenge still remains on how to derive a symmetry test with a good power performance and at the same time delivering a reliable Type I Error estimate. To overcome this problem, a new data-driven m -out-of- n bootstrap method is introduced for testing symmetry about an unknown median. The asymptotic properties of the developed m -out-of- n bootstrap tests are investigated along with their empirical finite-sample performance. The new tests are illustrated by applications to legal studies and wildlife monitoring.

A conversation about implicit bias

Explicit bias reflects our perceptions at a conscious level. In contrast, implicit bias is unintentional and operates at a level below our conscious awareness. Implicit stereotypes shaping implicit biases are widely studied in criminal justice, medicine, CEO selection at Fortune 500 companies, etc. However, the problem of unconscious bias remains. E.g., while women constitute an increasing proportion of all STEM undergraduates, they still make up only a small proportion of faculty members at research universities, and they are substantially under-represented in organizational leadership and as recipients of professional awards and prizes. Can we afford to have unintentional perceptions continue to hinder the success and advancement of women and other underrepresented groups? Can we afford to continue to underuse human capital in science? This session at the 2015 Joint Statistical Meetings (JSM) aimed to illuminate what statisticians need to know and do to break the glass ceiling of implicit bias.

Catching Uncertainty of Wind: A Blend of Sieve Bootstrap and Regime Switching Models for Probabilistic Short-Term Forecasting of Wind Speed

Although clean and sustainable wind energy has long been recognized as one of the most attractive electric power sources, generation of wind power is still much easier than its integration into liberalized electricity markets. One of the key obstacles on the way of wider implementation of wind energy is its highly volatile and intermittent nature. This has boosted an interest in developing a fully probabilistic forecast of wind speed, aiming to assess a variety of related uncertainties. Nonetheless, most of the available methodology for constructing a future predictive density for wind speed are based on parametric distributional assumptions on the observed wind data, and such conditions are often too restrictive and infeasible in practice. In this paper we propose a new nonparametric data-driven approach to probabilistic wind speed forecasting, adaptively combining sieve bootstrap and regime switching models. Our new bootstrapped regime switching (BRS) model delivers highly competitive, sharp and calibrated ensembles of wind speed forecasts, governed by various states of wind direction, and imposes minimal requirements on the observed wind data. The proposed methodology is illustrated by developing probabilistic wind speed forecasts for a site in the Washington State, USA.

A local factor nonparametric test for trend synchronism in multiple time series

The problem of identifying joint trend dynamics in multiple time series, i.e., testing whether two or more observed processes follow the same common trend, is essential in a wide spectrum of applications, from economics and finance to climate and environmental studies. However, most of the available tests for comparing multiple mean functions either deal with independent errors or are applicable only to a case of two time series, which constitutes a substantial limitation in many modern, typically high-dimensional, studies. In this paper we propose a new nonparametric test for synchronism of trends exhibited by multiple linear time series where the number of time series N can be large but fixed. The core idea of our new approach is based on employing the local regression test statistic, which allows to detect possibly non-monotonic nonlinear trends. The finite sample performance of the new synchronism test statistic is enhanced by a nonparametric hybrid bootstrap approach. The proposed methodology is illustrated by simulations and a case study on insurance claims due to extreme weather.

Using the bootstrap for statistical inference on random graphs

In this paper we propose a new nonparametric approach to network inference that may be viewed as a fusion of block sampling procedures for temporally and spatially dependent processes with the classical network methodology. We develop estimation and uncertainty quantification procedures for network mean degree using a “patchwork” sample and nonparametric bootstrap, under the assumption of unknown degree distribution. We provide a heuristic justification of asymptotic properties of the proposed “patchwork” sampling and present cross-validation methodology for selecting an optimal “patch” size. We validate the new “patchwork” bootstrap on simulated networks with short- and long-tailed mean degree distributions, and revisit the Erdos collaboration data to illustrate the proposed methodology. The Canadian Journal of Statistics xx: 1–22; 2015

Evaluating the Impact of Climate Change on Dynamics of House Insurance Claims

The adverse effects of climate change bring increasingly more alterations to all aspects of human life and welfare, and one of the sectors that is particularly affected by changing climate is the insurance sector. Indeed, the year 2013 brought a record number of claims and substantial losses due to weather-related damages, and in the USA and Canada alone, the extreme weather events cost the insurance industry more than 3 billion dollars. The objective of this paper is to provide statistical data-driven insight on the (non)linear relationship between weather-related house insurance claims and atmospheric variables and to predict future claim dynamics accounting for changes in extreme precipitation. In this paper we propose to employ a flexible Generalized Autoregressive Moving Average (GARMA) model for count time series of claims, develop a new method to compare tails of the observed and projected extreme precipitation, and evaluate the impact of climate change on a number of house insurance claims in the GARMA framework. We illustrate our approach by studying insurance dynamics in four Canadian cities.

Forecasting Bitcoin Price with Graph Chainlets.

Over the last couple of years, Bitcoin cryptocurrency and the Blockchain technology that forms the basis of Bitcoin have witnessed a flood of attention. In contrast to fiat currencies used worldwide, the Bitcoin distributed ledger is publicly available by design. This facilitates observing all financial interactions on the network, and analyzing how the network evolves in time. We introduce a novel concept of chainlets, or Bitcoin subgraphs, which allows us to evaluate the local topological structure of the Bitcoin graph over time. Furthermore, we assess the role of chainlets on Bitcoin price formation and dynamics. We investigate the predictive Granger causality of chainlets and identify certain types of chainlets that exhibit the highest predictive influence on Bitcoin price and investment risk.