[Next] [Previous] [Top] [Contents]

Artificially Intelligent Specification and Analysis of Context-Dependent Attribute Preferences

5 An Application


The IMIS system has been applied to a variety of markets in order to create CDAP models. One such application was to 28 monthly observations of scanner data covering supermarket sales of spirituous liquors in a large American conurbation. The data covered price and sales volumes by bottle of between 150 NS 200 different brands and bottle sizes. A United Distillers marketing expert decided upon the key attributes, measured on a Lickert scale, for each brand.

The IMIS system was given one focus brand from which it selected a competitive set of 4 other brands. The focus brand itself had a market small market share of less than five per cent on average in this competitive set although the brand itself has a significantly larger share worldwide. One issue for analysis is, therefore, why a generally successful brand is exceptionally unsuccessful in this one market.

The ACHE algorithm was set to work on the first 10 dates only in each of two experiments. In the first experiment, the algorithm settled on a model with a single preference state. In the second, we reduced the fitness discount associated with larger numbers of CDAP states and, as a result, identified four CDAP states. Each of these models was used to predict the future shares over the hold-out set. The model with a single CDAP state predicted market shares with an RMSE over all 5 brands of 7.1 per cent. The model with four CDAP states predicted the shares with an RMSE of 5.9 per cent. The graphs below exhibit the comparisons of the simulated and actual market shares of three of the brands. The other two brands accounted for less than 4% of the market and the models both ignored those brands by predicting zero market shares for them.

Increasing the number of CDAP states increases the accuracy of the system's simulation of market shares over whole data set. It also gives a very picture of the demand side of the market. This difference is discussed in section 5.3.

5.1. - The single-CDAP-state model
5.2. - The four-CDAP-state model
5.3. - A comparison of the two models

Artificially Intelligent Specification and Analysis of Context-Dependent Attribute Preferences - 03 NOV 97
[Next] [Previous] [Top] [Contents]

Generated with CERN WebMaker