TRL
TOP PAGETokyo Research LaboratoryEmploymentProjectsRelated InformationIBM Research
Japanese page is here.  

Time series forecasting

Forecasting ??


Overview

We are exploring statistical methods for forecasting future behaviors of time series. Recently information technologies and optimization algorithms are aggressively applied to various problems in the area of logistics and supply chain management. Accurate forecasting of demand change often improves the quality of solution to such resource planning problems as production planning and replenishment planning.

Research items

  • Time series data analysis
    We analyze time series based on state space model. Parameters in model and state variables are estimated by maximum likelihood method with numerical optimization or simulation.
  • Huge time series data
    Dealing with huge time series has a lot of practical importance. We expect improvement of accuracy of forecasting by aggregating several analogous time series and distributing the forecast at the root level to leaf level. Therefore clustering of time series is a key technology to deal with huge time series effectively.
  • Special time series data
    State space model is very general, and even Kalman filter, which is a linear and Gaussian family of the model, can deal with various cases. However, we need to design a special model to achieve better accuracy of the forecasting.

Related information



Research home IBM home Order Privacy Legal Contact IBM
Last modified 30 October1999