Automated Modeling System
More than a year in the making, Numerical Alchemy has developed a patent pending automation modeling system that fundamentally changes the cost, speed, and accuracy of predictive analytics. Our pricing is competitive with companies using offshore resources, but you get results in just days (sometimes in just one or two) versus the 2 to 4 weeks it takes most of our competitors. But what about accuracy? The accuracy of our system is comparable to custom made models developed with the latest data mining techniques. In addition, the system automatically generates profiles of the important predictors so you have a better idea about the characteristics of likely churners or purchasers. Want to know more? Then contact Numerical Alchemy for more details on this exciting new technology.
More than a year in the making, Numerical Alchemy has developed a patent pending automation modeling system that fundamentally changes the cost, speed, and accuracy of predictive analytics. Our pricing is competitive with companies using offshore resources, but you get results in just days (sometimes in just one or two) versus the 2 to 4 weeks it takes most of our competitors. But what about accuracy? The accuracy of our system is comparable to custom made models developed with the latest data mining techniques. In addition, the system automatically generates profiles of the important predictors so you have a better idea about the characteristics of likely churners or purchasers. Want to know more? Then contact Numerical Alchemy for more details on this exciting new technology.
© copyright 2007 Numerical Alchemy, Inc.
Data is read into the system along with specifications for the dependent and potential predictor variables.
Data cleaning transforms categorical variables, imputes missing values, trims extreme values, and redistributes skewed distributions.
Fast variable selection routines zero in on predictors with promising predictive power.
The automated model building facility fits a variety of spline basis transforamtions to capture non-linear relationships. The types of splines that are fit include:
- b-splines
- linear basis functions
- natural cubic splins
- radial basis functions
- step functions
Model performance is assessed on the development and hold out validation samples.
The predictor variables that make it into the final model are then profiled in order to show the characteristics of those with higher scores vs. everyone else.
Scoring code is generated containing variable transformation logic and the final predictive scoring algorithm .
Numerical Alchemy's Automated Modeling System
Numerical Alchemy's Automated Modeling System
Segmentation Workbench
Like our automated modeling system, Numerical Alchemy's segmentation workbench creates and tests a variety of clustering solutions in order to quickly produce the optimal segmentation scheme on a set of data. Our workbench handles both categorical and continuous variables. Specifically, our system:
Like our automated modeling system, Numerical Alchemy's segmentation workbench creates and tests a variety of clustering solutions in order to quickly produce the optimal segmentation scheme on a set of data. Our workbench handles both categorical and continuous variables. Specifically, our system:
- Cleans and transforms variables prior to analysis
- Orthogonalizes data to eliminate correlations between attributes
- Eliminates outliers via fast multi-dimensional outlier detection
- Produces a range of clustering solutions using a variety of techniques including K-means, recursive divisive methods, and hybrid clustering approaches
- Various solutions are assessed on a variety of statistics like CCC, F-tests, incremental R-square values
- Variables are then profiled on the most promising solutions for further interpretation and assessment