Nuggets uses an exciting and new information technology SiftAgent™which incorporates state-of-the-art proprietary computer algorithms.
The software agents can quickly search vast amounts of data for patterns. These rules in the form of "If Then" statements are stored in a rule library which is the "model".
New data can then be presented to the model which finds the best applicable rule for each instance. These rules are the predictions.
Nuggets is a complete modeling system with the ability to not only build models but to validate them against holdout sets and predict against new data.
There are many functions that allow you to understand the underlying patterns
There are no arcane mathematical formulas which hide the patterns and are often not understandable to management.
Technical and non-technical decision-makers can use these rules for various mission critical purposes, such as increasing top line revenue, managing unit costs, developing niche-specific marketing campaigns, analyzing medical data, evaluating insurance underwriting risk, and much more.
Why is our SiftAgent™ technology used in our Nuggets® products better?
1. The data that applies to business and scientific problems often doesn’t comply with the requirements of the current group of Data Mining methods. Some example of these requirements and the way Nuggets overcomes them are:
- Small number of predictor variables often less than 10 or 15.
- Nuggets can handle tens of thousands without attribute reduction methods
- Difficult to deal with mix of nominal and numeric characters
- Nuggets can handle both
- Predictor attributes must be uncorrelated
- Correlation among predictor attributes is allowed with Nuggets
- Missing values not allowed
- Nuggets uses all data and ignores missing values. Noise generating imputation is avoided.
- Non – linearities not allowed
- Not an issue with Nuggets
- Small patterns lost
- Small patterns are not lost. Nuggets will generate rules for these
- Need statistical and/or other specialized knowledge
- Nuggets can be used by non-technical people. Knowledge of the problem and data is still required.
- Numerical values need to be either linearized or have other transformations applied
- Nuggets can use numerical values as they are given without user supplied transformations.
- Building models are slow handcrafting exercises
- Nuggets has a built-in "Autobuild" module that enables models to be built in minutes with a few keystrokes. Of course you can also use many control parameters in the "Manual Build" mode if you wish to tweak the model.
- Accuracy is often low
- Because of the lack of data restrictions in Nuggets, the accuracy of predictions are enhanced