Building Predictions
...With Confidence
Predicting with confidence is a powerful tool for any decision making process. To establish this level of confidence, the rules used to establish the prediction must be completely transparent and descriptive to the end user.
The data mining approach exposes patterns and interprets them as easily understood rules describing the relationships between system variables. Exposing such hidden rules allows domain experts to make confident, informed decisions about why the detected patterns occur and when they are likely to occur in the future.
Once we have uncovered all of the significant patterns in a data set, we can form a rule model for making accurate predictions about how a system will behave. To make a prediction, an analyst selects a dependent variable to predict and any number of independent variables that are likely to affect the outcome. A search through the entire rule model reveals which patterns significantly impact the dependent variable, while assigning a Weight of Evidence for each pattern indicating how likely it is to contribute to the target event.
We can also classify new data using the same approach, adding together the combined evidence for each rule to support the results.
