Presenter/Author Information

Dimitri P. Solomatine

Keywords

local models, modular modelling, committees, neural network, flood forecasting

Start Date

1-7-2006 12:00 AM

Description

Data-driven models based on the methods of machine learning have proven to be accurate tools in predicting various natural phenomena. Their accuracy can be however increased if several learning models are combined in an ensemble or a committee. Modular model is a particular type of a committee machine and is comprised of a set of specialized (local) models each of which is responsible for a particular region of the input space, and may be trained on a subset of the training set. This paper presents a number of approaches to building modular models. An issue of including a domain expert into the modelling process is also discussed, and the new algorithms in the class of model trees (piece-wise linear modular regression models) are presented. Comparison of the algorithms based on modular local modelling to the more traditional “global” learning models shows their higher accuracy and the transparency of the resulting models.

Share

COinS
 
Jul 1st, 12:00 AM

Optimal modularization of learning models in forecasting environmental variables

Data-driven models based on the methods of machine learning have proven to be accurate tools in predicting various natural phenomena. Their accuracy can be however increased if several learning models are combined in an ensemble or a committee. Modular model is a particular type of a committee machine and is comprised of a set of specialized (local) models each of which is responsible for a particular region of the input space, and may be trained on a subset of the training set. This paper presents a number of approaches to building modular models. An issue of including a domain expert into the modelling process is also discussed, and the new algorithms in the class of model trees (piece-wise linear modular regression models) are presented. Comparison of the algorithms based on modular local modelling to the more traditional “global” learning models shows their higher accuracy and the transparency of the resulting models.