Basic natural language details based on insights

Overview

Details that are based on insights also provide insights from related visualizations that are informative and easily understood in the context of the current chart. This allows for more comprehensive details that are related to the current visualization.

Algorithms

Details that are based on insights directly use the full scope of computations and statistical tests that are supported by the insights. Details also obtain the results for related visualizations and compile them together in a meaningful message. Details also provide some additional analysis based on the insights output by producing further details suitable for textual output.

Details

Single explanatory field

Given a response field and a categorical explanatory field, details use insights to detect relationship between the response and the explanatory field. Analysis is applied depending on whether the response summarization level is average or count and the predictive strength is reported if a relevant relationship is discovered.

If the explanatory field is numeric and insights generates a fit line, then IBM® Cognos Analytics reports positive or negative slope for linear fit and point out whether a quadratic relationship is detected. If the relationship is quadratic, Cognos Analytics also reports the extremum point. Cognos Analytics computes the minimum or maximum value of the response and the explanatory value where the response extremum occurs.

Two explanatory fields

When two explanatory fields are available, Cognos Analytics detects relationship between the response and both explanatory fields and the relationship between the response and each explanatory field separately. If the predictive strength for the relationship with both explanatory provides relative improvement greater than 10% over each separate relationship, Cognos Analytics reports that the response is affected by both explanatory fields. Otherwise, it is affected by a single explanatory field, or by each explanatory field separately, but not together.

Meaningful differences

Details reports all meaningful differences that are discovered by the insights for the used data when response summarization level is count, average, or sum. They are reported for categories of each explanatory field separately as well as for the combinations of categories for two explanatory fields. Indications whether values are unusually high or low are also provided.

Decision trees

Details report predictive strength for a decision tree and a list of predictor fields that are used for splitting the tree nodes. A predictor field with the highest variable importance and its ratio improvement of variable importance over other fields in the decision trees is reported if the ratio is greater than two. A similar detail can be displayed for a field with the lowest variable importance.