时间：2019年03月13（星期三）14：40-15：40

地点：学院南路校区，学术会堂706

报告题目：Feature-basedTime Series Forecasting

报告人：Thiyanga Talagala，Monash University

报告摘要：This workpresents three feature-based algorithms for large-scale time series forecasting.The algorithms aredeveloped based on meta-learning approach. Inour first algorithm we use a random forest algorithm to identifythe bestforecasting model. We call this framework FFORMS (Feature-based FORecast ModelSelection).Inthe second algorithm, FFORMA (Feature-based FORecast Model Averaging), we usegradient boostingto obtain the weights for forecastcombinations. The third algorithm use efficient Bayesian multivariatesurfaceregression approach to estimate forecast error for each method, and then usingthe minimum predictederror to select a forecasting model or tochoose individual models for forecast combinations. The proposedalgorithmsperform well compared to several benchmarks and other commonly used approachesin large-scaleforecasting.

报告人简介：Thiyanga Talagala graduates from Monash University,Australia in 2019. Her PhD research focuses on the problem of forecasting largecollections of time series. Her research interests include time seriesanalysis, applied statistics and statistical computing.

本次活动受中央财经大学2019专题学术讲座项目资助。