基于误判代价的C5.0算法的优化分析

针对决策树C5.0算法在建模中不同代价值的错误分类没有在建模过程中区别对待,使得模型错误分类代价较高的问题.论文使用误判代价值和代价矩阵以降低高代价错误率,从而实现在模型总体错误率变化不大的情况下,实现C5.0算法所建模型的错误分类代价最小.实验证明优化后的模型在测试数据中高代价错误率从原模型的1.52%降到了0,说明代价矩阵的应用效果非常明显,一般代价错误率也有所下降,低代价错误率基本持平.

The Optimization of C5.0 Algorithm Based on Misclassification Cost

According to the classification error C5.0 algorithm of decision tree mining model in different generation values in the data is not in the process of modeling the distinction, which makes the model error classification cost higher problem. In this paper, using the cost ma- trix and error value by setting to reduce the high cost of error rate, so as to realize the overall er- ror rate in the model do not change much, the realization of C5.0 algorithm the model misclassifi- cation cost minimum, experiments found that the optimized model in testing high cost data error rate from the original model. 1.52% down to 0, the application effect of cost matrix is very obvi- ous, the general price error rate also fell, low error rate was essentially flat.