A Cost-sensitive Decision Tree under the Condition of Multiple Classes

Cost-sensitive learning is one of the top ten problems in the field of data mining, its target is to produce the least cost in the classification process under the condition of achieve a given classification accuracy. The decision tree is a kind of cost-sensitive classification algorithm. There are many typically cost-sensitive decision tree algorithms based on greedy method to build a single model such as PM, MinCost, etc. This kind of algorithms has good comprehensibility, requires less time and space complexity compared to other cost-sensitive classification algorithm. However,, currently a lot of research works are limited to binary classification problem, very few people study the performance of classification cost and accuracy of this kind of algorithm under the condition of multiple classes. This paper puts forward a cost-sensitive decision tree based on score-evaluation under the condition of multiple classes (SECSDT_ MC for short). Experiments show that SECSDT_ MC compared with PM and MinCost can produce fewer classification costs or achieve higher classification accuracy in most cases under the condition of multiple classes.