Media.Player

A Novel System for Predictive Modeling and Analysis

Technology Benefits

- Create simple and easy to interpret models- Provide more accurate prediction models- Use only a small number of patterns and local models

Technology Application

- Data mining- Predictive model construction, for regression, classification, and ranking- Predictive model analysis, for regression, classification, and ranking- Identifying diverse groups of data requiring different treatments

Detailed Technology Description

None

*Abstract

The inventors created a new modeling system called Contrast Pattern Aided Prediction Modeling (CPXR) for performing predictive modeling and analysis. CPXR builds accurate new prediction models from existing data. It also identifies data groups where large prediction errors occur when using an existing prediction model and provides one or more local prediction models to correct these large prediction errors. CPXR is especially effective in the case of high dimensional data that often contains diverse subgroups with highly different predictor-response relationships. Using 50 real data sets, the performance of CPXR was compared with other well-known regression methods, such as Linear Regression (LR), Piecewise Linear Regression (PLR), Support Vector Regression (SVR), Bayesian Additive Regression Trees (BART), and Gradient Boosting (GBM). CPXR outperformed them significantly on average regarding accuracy, overfitting, and sensitive to noise.

*Principal Investigator

Name: Guozhu Dong, Professor

Department: Computer Science

Name: Vahid Taslimitehrani

Department:

Country/Region

USA