On this page, you can create or modify an algorithm. If you are modifying an algorithm, current values are entered automatically, and you modify one or more values. If you are creating an algorithm, you enter all required values.
In the left pane of the Enter Algorithm and Accessor Information page, you select one of the following algorithms:
Association rules—Provides market basket analysis. Results are reported in the following form: nn% of users who buy Product X buy Product Y . For example, 80% of the people who buy root beer buy ice cream.
Clustering—Recognizes patterns and arranges items into groups. Clusters can provide information about market segmentation and can be used with various predictive tools. For example, clusters can predict the kinds of users that are most likely to respond to root beer ads.
Decision tree—Creates decision trees that can be used for classification and prediction. The algorithm results are the answers to a series of yes and no questions—A yes answer leads to one part of the tree, and a no answer leads to another part of the tree. For example, a decision tree can tell you to suggest ice cream to a customer who buys root beer.
Naive Bayes—Calculates conditional probabilities of outcome by examining the number of elements that occur with the target values. Naive Bayes algorithms assume that all elements are statistically independent. Naive Bayes generates a quick result. Therefore, you may want to use it before you use the clustering or decision-tree algorithms.
Neural net—Constructs a multi-layered set of nodes. The nodes of one layer connect to the nodes of the next layer and, thus, create a net that resembles human neurons (a net that models numerical dependencies between input and output data). Neural network algorithms generalize and learn from data. For example, neural networks can be used to predict financial results.
Regression—Identifies dependencies between one value and other values. For example, multilinear regression can determine how the amount of money spent on advertising and payroll affects sales values.
After you select an algorithm, in the following fields, you enter or modify mining task settings, parameter values, and accessor values:
Mining Task Settings—You use this option to determine how the data-mining framework and the algorithm handle missing values. If you want a non-default setting, you must set the option each time you run the mining wizard.
Parameter Values—Most algorithms have a set of parameters that is specific to the algorithm. For parameter value requirements, see algorithm-specific information.
Accessor Values—Accessor values, which can be predictors or targets, determine the set of data to mine. The predictor and target requirements are defined by the algorithm. When you select an accessor node (in the left subpane), attribute domain fields to be completed are displayed in the right subpane.
Note: | Target values for some attributes are provided automatically, with the associated predictor values. |