Data Mining & Analysis
Examples
Example #1
Stochastic algorithms for generating synthetic time series of weather elements provide useful input for a variety of agricultural and hydrological models. Synthetic multivariate time series have traditionally been limited to multiple weather elements as a specific site. In this study a multivariate generator of daily average temperatures at 148 U.S. surface stations is developed. The goal is to provide multi-station time series with temperature fluctuations that closely replicate the autocorrelation and spatial covariance of the observed time series.
(Click on images for full size.)
Example #2
The recent price action of commodities may hold clues for subsequent price evolution. A self-organizing feature mapping procedure is used to postion historical trading days onto a closed 2-dimensional surface map (SOFM) such that nearby trading days share similar features. Features are defined, a priori, from each day's recent price action. The features of a recent trading day similarly determine its position on the map. Its subsequent price evolution can be probabilistically estimated from the "neighboring" historical trading days' price evolutions. These neighboring trading days are referred to as historical analogs (HAs). The following images from the analysis pertain to projected 15-day trends in Gold prices on June 9, 2010.
(Click on images for full size.)
Other Examples
CCC has used data mining techniques to extract meaningful relationships between weather elements and revenues from cement sales. Aspects of this project are provided under the Business Weather Sensitivity link.
Data mining techniques were used to sift through numerous fog cases to classify them into (1) regional types, with little influence from the local industrial water vapor emissions, and (2) localized types, with stronger dependency on industrial water vapor emissions. Aspects of this project are described under the Custom Projects link.