Mixture Models for Exploratory Data Analysis
Click to view larger
Description:The Naval Research Laboratory (NRL) has developed a method to automatically discover and evaluate candidate mathematical models within unstructured data and classify each data point with a probability of fitting each discovered model. A model may be any mathematical shape, including closed curves and geometric shapes, general functions such as Gaussians, polynomials, exponentials, or transcendentals, or any combination of the above. Once a set of models is determined to fit the data set, a probability may be assigned to each data point to measure the likelihood that a given point belongs to a particular function, i.e. that it is consistent with that model. A classification of these points may be derived from these probabilities. This analytical tool is intended for data mining and visual analysis of large, complex data sets in the intelligence community. It has been used with GIS data, weather data, crime data, and other multivariate data sets as well. This software is coded in C++ and is available for commercialization.
- Model abstraction seamlessly considers different types of functions as equally possible candidates
- Does not require an estimate of the number of underlying mathematical models
- Allows data points to be probabilistically classified into multiple models or identified as outliers
- Includes a few parameters that the analyst can use to adjust the output
- Intelligence analysis
- Research & education
- Economics, finance, real estate, and business analysis
- Targeted, demographics-based marketing
Click here for a more detailed PDF version of this technology.
Contact:Naval Research Laboratory
Technology Transfer Office, Code 1004