Strategy ONE

Model tab, View sub-tab

The View sub-tab contains a visual description of the model. Its behavior depends on the type of model.

Regression and General Regression Models

For Regression and General Regression models, the regression equation is shown. There is a check box that adds parentheses that can make the equation easier to understand. There is also a One Term per Line check box that formats the equation with a single variable per line.

Neural Network Models

For Neural Network models, the neurons and connections that make up the model are shown, including all the information about neurons (activation function and its parameters), connections and their weights, and input/output transformations.

Tree Models

For Tree models, each node that makes up the tree is shown. All information associated with the node is visible, including the score represented by the node, the predicate used to evaluate the node, and the distribution of scores used when training the model. It also includes an Eye chart, where the outer ring shows the total population distribution and the inner pie chart shows the distribution for that particular node.

One of the following tree styles can be selected:

  • Large-Vertical: This is a decision tree with the root node at the top and children below. All information about each tree node is displayed.

  • Large-Horizontal: This is a decision tree with the root node at the left and children to the right. All information about each tree node is displayed.

  • Compact-Vertical: This is a decision tree with the root node at the top and children below. For each node in a standard tree, only an eye chart is shown and, if available, the percent of the population represented by this node. For each node in a tree with regression nodes, only a bar graph is shown and, if available, the percent of the population represented by this node.

  • Compact-Horizontal: This is a decision tree with the root node at the left and children to the right. For each node in a standard tree, only an eye chart is shown and, if available, the percent of the population represented by this node. For each node in a tree with regression nodes, only a bar graph is shown and, if available, the percent of the population represented by this node.

  • List: This representation has a simple tree view on the left and details about each node on the right. Selecting a tree node on the left will highlight the information about that node on the right. If the model is too large to be displayed, this tree style becomes the only option for viewing the decision tree.

  • Rules: This representation displays each potential outcome of the predictive metric, including the rules that result in each potential outcome.

Cluster Models

For Cluster models, a table describing each cluster is provided. This information, especially in conjunction with a study of model response using the Simulator, can provide a good understanding of the clusters in the model (see the Simulation tab).

Mining Models (model selection and sequencing)

For Mining models which represent a sequence of models, each sub-model in the sequence is displayed on an individual tab. The first tab provides a summary of the entire mining model, detailing how the results of each sub-model are related to sub-models which appear later in the sequence. Following the summary tab is one tab for each sub-model, listed in the same order as the sub-models appear in the mining model.

Support Vector Machine and RuleSet Models

These types of models currently cannot be displayed.

Association Rule Models

The rules contained in the Association rules model are displayed in a table. The rows in the table represent the rules' antecedents and the columns represent the consequents. The following options in the lower left select the statistic to display in the body of the table:

  • Confidence: An estimate of the probability of a transaction having the consequent given the antecedent.

  • Support: The relative frequency of transactions containing both the antecedent and the consequent.

  • Lift: A ratio that describes whether the rule is more or less significant than random chance. Lift values greater than 1.0 indicate that transactions containing the antecedent tend to contain the consequent more often than transactions that do not contain the antecedent.

  • Leverage: A value that describes the support of the combination of the antecedent and the consequent as compared to their individual support. Leverage can range from -0.25 to 0.25, and a high leverage indicates that there is a relationship between the antecedent and the consequent. For example, if 50% of the transactions contain the antecedent and 50% of the transactions contain the consequent, you would expect 25% of the transactions to contain both the antecedent and the consequent if they were completely independent; this would correspond to a leverage of zero. If more than 25% of the transactions contain the antecedent and consequent together, then there is a positive leverage (between 0 and 0.25). This positive leverage indicates that the antecedent and consequent appear more frequently than you would expect if they were completely independent, and can hint at a relationship.

  • Affinity: A measure of the similarity between the antecedent and consequent itemsets, which is referred to as the Jaccard Similarity in statistical analysis. Affinity can range from 0 to 1, with itemsets that are similar approaching the value of 1.

The legend at the bottom of the table displays the minimum and maximum values of the selected statistic in the model including the color used to represent those values in the table.

Time Series Models

Information about the Time series model's implementation and key parameters are displayed.

These models are equations that represent the best profile that matches the data used to train the model. Each profile can be described by two aspects, trend and seasonality.

Trend can be one of the following types:

  • None

  • Additive

  • Damped additive

  • Multiplicative

  • Damped multiplicative

  • Triple exponential

The damped additive and damped multiplicative trends use the damping function, which constrains the trend from progressing indefinitely without change. The rate of this damping is governed by the damping parameter φ (phi). The value of phi can vary from zero to one, with the damping effect becoming smaller as phi approaches one.

Seasonality may or may not be present in the model. If it is present, it is either added to the trend equation (additive seasonality) or multiplied (multiplicative seasonality).

For more information on how trend and seasonality affect time series analysis, including example trend lines and formulas, see the Advanced Reporting Help.

Model View

For large Tree or Neural Network models, a window appears that shows the "bird's eye view" of the model. This window can be resized and docked by dragging its edges or using its minimize/maximize buttons. The window can be closed by clearing the check box on the lower left or clicking the window's "X" button. Selecting the check box makes the window reappear.