We propose a novel decision tree algorithm that can be used to lower the cost of feature computation while maintaining a high level of classification accuracy. The method determines the branching of each node by using a criterion that integrates the impurity of the data set and the expected computational cost. This approach enables data to be classified quickly and accurately. We also discuss its application to video analysis of shot boundary detection. The experimental results show that the proposed method detected shot boundaries and had a lower computational cost while maintaining the same accuracy as conventional algorithms such as the usual decision tree and support vector machine. Recall and precision were 96% and 90%, respectively, and the processing time was reduced by nearly half compared with that of the conventional algorithm.
