DM Stat-1 Software
Link to Home

Link to Articles

Link to Consulting

Link to Seminar

Link to Stat-Chat

Link to Software

Link to Clients

DM STAT-1 CONSULTING's flagship GenIQ Model Software
is an  implementation of the GenIQ Model©, which  explicitly maximizes expected response or profit – as displayed in a decile table – from direct/database marketing solicitations, telemarketing programs, CRM campaigns, Web broadcasts, and the like. The GenIQ Model is an assumption-free, nonparametric machine learning model based on Darwin's Principle of Survival of the Fittest, and natural genetic operations - namely, genetic programming. The genetic GenIQ-Response and GenIQ-Profit modules offer a clear advantage over logistic and ordinary regression methods, whose performance is dependent on theoretical assumptions, a pre-specified parametric model, and data restrictions. Pointedly, the GenIQ Model automatically determines the best set of predictor variables (from the original variables, and newly constructed – genetically data mined – variables) based on a virtually unbiased assessment of all variables under consideration, an achievement not possible with statistical methods.

Be among your cutting-edge colleagues, who have expressed their nonrandom words of praise for the GenIQ Model. I guarantee GenIQ will make you rethink regression modeling forever; if not, I will give you a free copy of  my latest  book Statistical Modeling and Analysis for Database Marketing: Effective Techniques for Mining Big Data. Experience for yourself the power of the GenIQ Model - to any of your past or present modeling projects. Request the full-featured GenIQ Model Software to rebuild or build your own GenIQ model to compare with your past models, or to assess the ease-of-use and predictiveness of the GenIQ Model on your current project. (Paradoxically, the less you know about statistics, the better candidate  you are for GenIQ.) If you prefer, provide me with the datasets for building and validating one of your models. After employing GenIQ to your data, I will provide you a complete paper-trail of the procedure and report the new model results. If you're interested yourself in perhaps the easiest and most predictive method of building either a database response or profit model, click here.

Heeding the need of its target marketplace, the developers of the primo predictive analytics tool – the GenIQ Model Software – have added an Excel© toolbar. Predictive Analytics Now Accessible to Excel Spreadsheet Users

DM STAT-1 CONSULTING, the leading firm for analysis and modeling in the DM Space (direct/database marketing, CRM, and data mining/knowledge discovery), and a premier developer of software tools that rely on machine learning (ML) technologies, such as genetic programming, have collaborated in the development of powerful, yet user-friendly applications of hybrid statistics-ML models.

The new partnership DM STAT-1 STATWARE has refined the statistical models used in DM to make them more powerful. Thanks to the flexibility of ML computation, our new models explicitly address DM objectives - a major improvement over their traditional counterparts. The new models are efficiently estimated with algorithms that harness the power of the desktop computer.

The GenIQ Model can comfortably accomondate BIG data consisting of hundreds of thousands of observations. For BIGGER data consisting of a million or more observations, DM STAT-1 STATWARE has an arrangement with Dr. John R. Koza of Stanford University (the inventor of genetic programming) to implement the GenIQ Model on his Cluster Computer System comprised of 1,000 Pentium II 350-MHz personal computers.

Database Marketing Analysis, Mining and Modeling
GenIQ is an automatic model generation tool that can significantly improve the response rates (from any binary target variable) and profitability (from any continuous target variable) of DM solicitations. Using genetic modeling techniques, GenIQ concentrates responders and high profit customers into the upper deciles of its model scores.

GenIQ addresses the primary goals in the DM Space by focusing on aggregates instead of individual scores to obtain (in the upper deciles):
  • as many responses as possible, and
  • as much profit as possible.
By explicitly maximizing the cumulative lift in the upper deciles, GenIQ builds high-performance response and profit models that significantly outperform other methods. In contrast, linear and logistic regression, discriminant analysis and neural nets achieve marketing goals indirectly. Rather than maximizing decile response and profit, other methods minimize squared error and derive scores that are not optimized for the problem. Recent trials show that GenIQ can boost cumulative lift in the top two to four deciles by 10% to 25% over traditional models. 

What is Genetic Modeling?
Genetic modeling is based on the Darwinian ideas of "survival of the fittest" and the natural genetic operators of reproduction (copying), mating (crossover), and mutation (random alteration). The process begins with a fitness function (in GenIQ, populating the upper deciles with as much response or profit as possible) and a set of user-selectable mathematical and logical operators. A first generation of as many as 250 - 1,000 models is randomly generated using the operators; the "fitness" of each model is evaluated using training data.

A second generation of models is then created through mating, reproduction, and mutation. When two models (parents) "mate" the offspring (children)  are mixtures of the parents; thusly, each parent probabilistically contributes good genetic material to the child. The frequency with which a model mates, is copied, or is altered is a function of its fitness score - how well it fills the upper deciles appropriately. After a suitable number of generations (typically 50 - 100), the forces of natural selection yield the best-of-generation model, superbly adapted to the model objective (fitness function).

For a technical discussion of genetic modeling (programming) click here

Graphical User Interface
GenIQ guides the user through all phases of importing data and setting up the model. The model parameters - population size, genetic functions, and breeding (reproduction, crossover, mutation) characteristics - are set to easily modified intelligent defaults.

Model evolution is fully automatic and can be paused at any time to assess the best-of-generation solution, alter breeding characteristics or change the genetic functions. Graphical displays summarize characteristics of the fully-evolved model, which can be easily exported to C, SAS, SPSS, Visual Basic, or Microsoft Excel for deployment.

Although the fitness functions in GenIQ optimize direct marketing outcomes, GenIQ can be used for a much broader class of problems. GenIQ is a tool of choice whenever concentrating specific records into well-defined groups is needed. For example, in identifying fraud, risky credit prospects, customers most likely to go bankrupt, or home owners most likely to refinance their mortgages, the objectives would be well served by the GenIQ objective. Advantages of GenIQ include:
  • Automatic variable selection
  • Automatic model building
  • A selection of alternative models to choose from
  • An option to favor smaller models
  • Automatic testing on a holdout sample
  • Export of any model to various source code formats (including SAS)
To run GenIQ a user only needs to specify a training database and select a target variable. Using broadly applicable defaults, GenIQ will do everything else: reserve a fraction of the data for testing, select a set of model construction operators, set breeding parameters, determine how many generations to evolve, and then run. When the run is complete (or any time during the evolution) the analyst can examine any of the models of the current generation in graphical source code form.

Experienced users will want to experiment with different program settings. For example, GenIQ provides an extensive collection of functions and operators from which models are built up, including arithmetic, trigonometric, logical, and fuzzy functions.

By default, only a core subset of these are made available for model evolution, but knowledge of the subject matter may suggest that other functions be allowed. The user can also define custom functions to add to the stock of  "genetic material."

System Requirements
  • Windows 98SE, Windows ME, Windows 2000 Windows XP, Vista 
  • Very latest processor recommended
  • At least 64MB of RAM

Technical Specifications

  • All input files formats, including SAS
  • Numeric and character variables allowed
  • Unlimited number of rows and columns
Go back to Home page.