I got my B.Comm.(Hons) from Department of Econometrics and Business Statistics, Monash University, Australia. My honours thesis focused on Exploratory visualisation of big time series data, supervised by Rob J Hyndman. My research interests lie in data visualisation, computational statistics and time series analysis.
Wanna talk? Please drop me a line to earo.wang(at)gmail.com
Nov 2013 – Present | Monash University
- Anomalous time series detection. I’ve proposed several statistical metrics to measure the “anomalousness” of time series and implemented the methods of outlier detection in the
anomalouspackage for R.
- R packages development on forecasting. I’ve rewritten the
htspackage in order to make it more efficient and flexible. I’ve also contributed some new functions to the
Jul 2014 – Jun 2015 | Monash University
- ETC1000 - Business and economic statistics - 2014 S2, 2015 S1.
- Hyndman, R. J., Wang, E., & Laptev, N. (2015). Large-scale unusual time series detection. Working paper. http://robjhyndman.com/working-papers/cikm2015/
- Hyndman, R. J., Lee, A., & Wang, E. (2014). Fast computation of reconciled forecasts for hierarchical and grouped time series. Working paper 17/14, Department of Econometrics & Business Statistics, Monash University. http://robjhyndman.com/working-papers/hgts/
Visualisation of big time series data
We examine some recently developed techniques for visualising multiple time series and propose some simple and quick methods for graphical representation of estimates of trend, seasonality and remainders from a time series. Furthermore, we also propose an approach to visualisation of very large collections of time series using some dimension reduction techniques. First, we compute a vector of statistical metrics describing a range of characteristics of each time series, and then use Principal Component Analysis to further reduce the data to two dimensions. The biplot, based on the first two principal components, provides a useful exploratory tool, and allows the identification of anomalous time series. We order the time series according to their data density and data depth in the space of the first two principal components. The method enables the most unusual series, based on their feature vectors, to be identified.
The R package hts presents functions to create, plot and forecast hierarchical and grouped time series. In forecasting hierarchical and grouped time series, the base methods implemented include ETS, ARIMA and the naive (random walk) models. Forecasts for grouped time series are calibrated using bottom-up and optimal combination methods. Forecasts for hierarchical time series are distributed in the hierarchy using bottom-up, top-down, middle-out and optimal combination methods. Three top-down methods are available: the two Gross-Sohl methods and the forecast-proportion approach of Hyndman, Ahmed, and Athanasopoulos (2011).
- Type: R package
- Title: Unusual Time Series Detection
- Description: Methods for detecting anomalous time series
- Link: Development Version
- BugReports: https://github.com/robjhyndman/anomalous-acm/issues