Companies that have successfully introduced SOM data mining to analyze survey data can, as the next step, introduce a method to simulate consumer behavior three or five years from now using a Bayesian network.


Each node of the Bayesian network corresponds to a question in the survey, and a network model is created by connecting the linkages between the questions with arcs (links). The CPT is similar to the cross-tables used in conventional survey analysis, and the Bayesian network can be regarded as a model of the entire survey data, with a collection of cross-tables. The Bayesian network can be regarded as a model of the entire survey data by means of a collection of cross-tables.


If the same survey is conducted regularly, the structure of the network will remain unchanged, and the parameters of the CPT will change with each survey. Therefore, by analyzing the time-series changes in the CPT, we can predict future CPT parameters based on the trends. Furthermore, it is possible to infer changes in the CPT by taking into account scenarios related to economic trends, social conditions, and technological trends.


Bayesian network software can learn from past case data to create CPTs, or, conversely, Bayesian networks with arbitrary values of CPTs can output case data for simulation. In other words, assuming a future CPT, the simulation can generate the survey results for that case. The resulting simulation data can then be reanalyzed using SOM data mining to determine future consumer behavior.

Quantitative Strategic Management