Selection of Time-Series for Clustering Supermarket Customers

		Pawan Lingras and Greg Adams

Selecting appropriate features to represent customers, for clustering, 
is necessary for establishing useful customer profiles in supermarket 
data mining. This paper reports results of experiments with 
time-series of six different criteria for clustering supermarket 
customers. Different combinations of criteria and weighting schemes 
resulted in similar general customer classes. However, the 
memberships in the resulting clusters and descriptions of customer 
profiles changed with the clustering schemes. The clustering was 
done using Kohonen neural networks. The results underscored the need 
for careful experimentation in determining the appropriate scheme.