Browse Prior Art Database

# Method and System for Forecasting Event Driven Online Traffic

IP.com Disclosure Number: IPCOM000201011D
Publication Date: 2010-Nov-04
Document File: 3 page(s) / 103K

## Publishing Venue

The IP.com Prior Art Database

## Related People

Konstantin Salomatin: INVENTOR [+4]

## Abstract

A method and system for forecasting event driven online traffic is disclosed. A Boosting Gaussian Process (BGP) is used for forecasting the event driven online traffic.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 52% of the total text.

Method and System for Forecasting Event Driven Online Traffic

Abstract

A method and system for forecasting event driven online traffic is disclosed.  A Boosting Gaussian Process (BGP) is used for forecasting the event driven online traffic.

Description

Disclosed is a method and system for forecasting event driven online traffic.  A Boosting Gaussian Process (BGP) is used for forecasting the event driven online traffic.

A Gaussian Process (GP) is generally used as a regression tool for modeling arbitrary smooth curve functions.  When compared to other modeling methods, a GP provides non-parametric models under Bayesian Framework and therefore, provides non-linear solutions and analytical predictions for sparse and non-uniform data.

In a GP, a stochastic process y( x) is a collection of random variables indexed by x  X such that values at any finite subset of X has a joint Gaussian distribution.  In an instance, X may be a set of time stamps, e.g.  X = {1,2,…,n}.  Forecasting functions y’(x) = f(x) are from a joint Gaussian distribution defined by:

f =(fl ,…,fn) ~ N(,Σ).                                                      (1)

The joint Gaussian distribution may be fully specified using a mean vector  and a covariance matrix Σ = [k(x,x’)] as,

f(x) ~ GP(,k(x,x' )),                         (2)

where  is an expectation of forecasts produced by all possible forecasting functions f at time x.  Further, covariance function k(x,x’) in equation (2) given by  is the covariance between forecasts at time x and all historical forecasts at time stamps x’.  The function k is a positive semi-definite function, also known as a kernel function.  Accordingly, for forecasting online traffic for a new point z, f(z) is calculated.  Fig. 1 illustrates the regression results of forecasting using a single GP.

Figure 1

The method and system disclosed herein uses a Boosting Gaussian Process (BGP) for forecasting the event driven online traffic.  BGP automatically partitions a signal containing a stream of data into multiple segments and thereafter, trains a GP for each segment.  Ideally, an event corresponds to one or more segments.  Therefore, traffic attributed to events and off-event traffic has associated global growth trends and stochastic characters.  BGP is a special case of Mixture of GPs (MGPs) and is similar to the Mixture of Experts approach.

Data stream is generally provided model parameters given by:

,

where GPj is the individual GP performed on each segment,  are parameters of gating network used for the multiple GPs and are discrete indicators that signify assignment of segments to each GP.

BGP uses a boosting tree tech...