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

Algorithm 2

CART algorithm.

Input:

D - ETC customer index dataset and their associated class labels;

minbucket - the minimum number of observations in any terminal (leaf) node.

Output:

A decision tree of ETC customer segmentation.

Method:

create a node N;

set a split point, a, for a specific segmentation index A, and split D into

subsets D₁ and D₂. Thus, for ETC segmentation index, three set of subsets

are obtained;

computerize the Gini indexes of three indexes in dataset D, respectively.

Determine an optimal splitting index;

repeat steps – until the samples in the subset are too few or the

reduction of “node impurity” cannot be below the given threshold and

create a leaf node;

the leaf node is labelled with the majority class in D to node N, and

generate a decision tree of ETC customer segmentation;

select different subtrees (branches) in the decision tree and prune it by the

cross-validated error and cost complexity;

output an optimal decision tree of ETC customer segmentation.