| |
by 
Product Description
Introducing MARS®, a new high speed predictive modelling solution that provides superior forecasting accuracy. An essential component of any comprehensive data mining solution, MARS automates the development and deployment of accurate and easy to understand regression models. Use MARS to build business intelligence solutions for problems such as predicting credit card holder balances, insurance claim losses, customer catalogue orders, and cell phone usage.
MARS is an innovative and flexible modelling tool that automates the building of accurate predictive models for continuous and binary dependent variables. Multivariate Adaptive Regression Splines was developed in the early 1990s by Jerry Friedman, a world renowned statistician and one of the co-developers of CART. Salford Systems' MARS, based on the original code, has been substantially enhanced with new features and capabilities in exclusive collaboration with Friedman.
MARS is an ideal data modelling tool when the analyst needs to both accurately predict a future outcome and to understand the "why" underlying the predictive model. For example, if the goal is to predict new credit card customers monthly charges on the basis of detailed credit bureau data, or how dollars spent on a high ticket consumer good vary with dollars spent on other products, MARS is capable of generating a highly accurate predictive equation. And, in addition to delivering predictive accuracy, MARS allows the analyst to more fully understand the underlying data patterns and relationships, thereby allowing him/her to tell a story and use these insights to make more strategic decisions.
MARS' models may be as simple as straight lines or as complex as multi-dimensional surfaces with cliffs, ridges, and sharp twists and turns. Whether the outcome the analyst is trying to predict has only a few drivers each with their own separate relationship or whether many factors interact in complex ways to determine the outcome, MARS is capable of discovering and representing this relationship in an accurate and understandable way.
Response modelling problems MARS can solve, for example, given binary (yes/no) response outcomes are: 1) will a homeowner refinance their mortgage in the next quarter? 2) will a household respond to a direct mail offer? or 3) will a bank customer sign up for a new credit card? MARS can also estimate the probability that a treatment for a medical condition will succeed or the probability that a policyholder will file a claim. MARS is also ideal for solving predictive modelling problems involving continuous outcomes such as:
- How much will a customer spend on their next catalogue order?
- How large a balance will a credit card holder carry?
- How many minutes will someone use the cell phone this month?
- In an insurance claim is filed, how much is the loss?
In addition to using MARS as a model building tool, data analysts use MARS as an exploratory tool to refine more conventional models (e.g., linear and logistic regression). By automatically detecting variable transformations and interactions, for example, MARS slashes the time required to build a logistic regression model by more than half and significantly improves the model's predictive accuracy. MARS can also be used in conjunction with decision trees to build high performance hybrid models. Successful MARS hybrids have been used to accurately predict whether a household will respond to a direct mail offer, refinance a mortgage, apply for a new credit card and a myriad of other marketing research challenges.
Principal Characteristics
MARS excels at finding optimal variable transformations and interactions, the complex data structure that often hides in high dimensional data. In doing so, this new generation approach to data mining uncovers business critical data patterns and relationships that are difficult, if not impossible, for other approaches to uncover.
Given a target variable and a set of candidate predictor variables, MARS automates all aspects of model development, including:
- Separating relevant from irrelevant predictors
Large numbers of variables are examined using efficient algorithms, and all promising variables are identified.
- Transforming predictor variables exhibiting a nonlinear relationship with the target variable
Every variable selected for entry into the model is repeatedly checked for non-linear response. Highly non-linear functions can be traced with precision via essentially piecewise regression.
- Determining interactions between predictor variables
MARS repeatedly searches through the interactions allowed by the analyst. Unlike recursive partitioning schemes, MARS models may be constrained to forbid interactions of certain types, thus allowing some variables to enter only as main effects, while allowing other variables to enter as interactions, but only with a specified subset of other variables.
- Handling missing values with new nested variable techniques
Certain variables are deemed to be meaningful (possibly non missing) in the model only if particular conditions are met (e.g., X has a meaningful non missing value only if categorical variable Y has a value in some range).
- Conducting extensive self tests to protect against over fitting
The user can choose to reserve a random subset of data for test, or use v-fold cross validation to tune the final model selection parameters.
MARS enables analysts to rapidly search through all possible models and to quickly identify the optimal solution, providing insights that can lead to a definitive competitive advantage. And, because the software can be exploited via an easy to use GUI, intelligent default settings, and aesthetically appealing output, for the first time analysts at all levels can easily access MARS' innovations.
MARS for Windows also incorporates two alternative control modes that extend the program's features and capabilities. In addition to controlling MARS with the GUI, you can also issue commands at the command prompt or submit a command file.
- User Friendly Graphical User Interface
MARS' easy to use GUI allows the user to control the variables and functional forms to be entered into the model and the interactions to be considered or forbidden, while allowing the MARS algorithm to optimize those parts of the model the analyst chooses to leave free. Once the model is selected, the user can easily remove or add terms, instantly see the impact of changes on model fit, review diagnostics that assist in model selection, save the model and apply the model to new data for prediction.
- MARS Output
MARS output is an easy to deploy regression model that can be automatically applied to new data from within MARS itself or exported as ready to run SAS® and C source code. To facilitate interpretation of the model, the output also includes interpretive summary reports as well as exportable two- and three dimensional curve and surface plots:
Data Translation Engine
The MARS data translation engine, DBMS/COPY®, supports data conversion -- both direct reading and writing -- of over 80 file formats, including:
- Statistical analysis packages: SAS® and SPSS
- Spreadsheets: Microsoft Excel and Lotus
Which version do you need?
MARS requires that all training data reside in RAM, so the larger the data set to be analyzed, the larger the RAM needed to analyze it. The exact amount of RAM required will vary from problem to problem. The table below is intended as a guide for the maximum number of candidate predictor variables that can be specified in a MARS analysis for the given sample size and amount of RAM workspace:
Number of Predictor Columns
You Can Use For Different Training Sample Sizes and MARS versions |
| Sample Size |
64 MB compile [2m]** |
128 MB compile [4.8m] |
256 MB compile [9.6m] |
512 MB compile [22.8m]*** |
| 10,000 |
200 |
480 |
960 |
2280 |
| 25,000 |
80 |
190 |
380 |
910 |
| 50,000 |
40 |
95 |
190 |
455 |
| 100,000 |
20 |
45 |
95 |
225 |
| 200,000 |
5 |
20 |
45 |
110 |
MARS run with default settings and with following assumptions: no missing values or categorical variables in training data; maximum interactions set to 1; maximum basis functions set to number of specified predictors.
NOTE that each variable containing a missing value counts as two predictors.
- ** Maximum number of numbers (in millions) based on above assumptions.
- *** Custom compiles up to 32 GB available on UNIX platforms. Maximum number of candidate predictor variables that can be specified regardless of available RAM is 8,192.
Increasing the Number of Variables MARS Can Handle
If you have a very large list of potential predictors, CART can be used first to extract the most important variables. MARS can then focus on the top variables from the CART model, enabling you to fit larger problem sizes into smaller workspaces and resulting in faster analyses and more accurate and robust models.
System Technical Requirements
- IBM PC 486 or higher
- 64 MB of RAM
- 10 MB of free hard disk space
- Windows 95/98/ME, Windows NT/2000/XP, Linux, or UNIX (Sun, SGI, HP, DEC, IBM)
Available Platforms
MARS is now available for Windows 95/98/ME, Windows NT/2000/XP, Linux and UNIX platforms, including DEC ALPHA, HP, Sun Solaris 2.5 and 2.6, SGI IRIX 6.2+ and IBM RS-6000.
Rule of Thumb for Calculating Required RAM
A rule of thumb that you can also use for calculating the needed RAM for your data set is to multiply the data set size by a factor of 3 to 4. For example, if your data set is 10 megabytes, MARS potentially requires 40 megabytes of RAM for the analysis.
© Copyright 2002 Salford-Systems Inc.
|
|
