Application of Customer Segmentation for Electronic Toll Collection: A Case Study

Algorithm 1

CLARA algorithm.

Input:

D - ETC customer index dataset;

k - the number of clusters;

samples - number of samples to be drawn from the dataset;

sampsize - number of observations in each sample.

Output:

The clustering results of ETC customer.

Method:

for i = 1 to samples, repeat (a)-(d);

(a) select sampsize objects randomly from ETC customer index dataset D as a sample, apply

the PAM algorithm to compute the best k-medoids – ;

(b) apply k-medoids to the entire dataset D and calculate the distance from every non-

medoids object in D to the closest object in the set , reassign each ETC

customer to different clusters;

(c) compute the average dissimilarity of this clustering, if the value is less than the current

minimum value, then replace the current value, and form the best k-medoids and the new

set of k representative objects;

(d) return to step , repeat the iterative process;

until no change, output clustering results of ETC customer.