Association Rules - Model Options

Use the settings on this tab to specify the scoring options for Association Rules models.

Model name You can generate the model name that is automatically based on the target field (or model type in cases where no such field is specified), or specify a custom name.

Maximum number of predictions Specify the maximum number of predictions that are included in the score result. This option is used with the Rule Criterion entries to produce the “top” predictions, where "top" indicates the highest level of confidence, support, lift, and so on.

Rule Criterion Select the measure that is used to determine the strength of the rules. Rules are sorted by the strength of criteria that are selected here in order to return the top predictions for an item set. You can choose from 5 different criteria.

• Confidence Confidence is the ratio of rule support to condition support. Of the items with the listed condition values, the percentage that has the predicted consequent values.
• Condition Support The proportion of items for which the conditions are true.
• Rule Support The proportion of items for which the entire rule, conditions, and predictions are true. Calculated by multiplying the Condition Support value by the Confidence value.
• Lift The ratio of rule confidence and the prior probability of having the prediction.
• Deployability A measure of what percentage of the training data satisfies the condition but not the prediction.

Allow repeat predictions To include multiple rules with the same prediction during scoring, select this check box. For example, selecting this enables the following rules to be scored.

bread & cheese -> wine
cheese & fruit -> wine

Only score rules when predictions are not present in the input To ensure that predictions are not also present in the input, select this option. For example, if the purpose of scoring is to make a home furniture product recommendation, then it is unlikely that input that already contains a dining room table is likely to purchase another. In such a case, select this option. However, if products are perishable or disposable (such as cheese, baby formula, or tissue), then rules where the consequent is already present in the input might be of value. In the latter case, the most useful option might be Score all rules.

Only score rules when predictions are present in the input To ensure that predictions are also present in the input, select this option. This approach is useful when you are attempting to gain insight into existing customers or transactions. For example, you might want to identify rules with the highest lift and then explore which customers fit these rules.

Score all rules To include all rules during scoring, regardless of the presence or absence of predictions, select this option.