基于下近似分布不变的乳腺癌诊断决策

Diagnosis Decision of Breast Cancer Based on Lower Approximation Distribution

指出变精度邻域粗糙集整体精度虽然较高,但个别决策类精度可能较低,不利于乳腺癌诊断。针对该问题,提出基于下近似分布不变的改进方法。并依据各属性的出现概率和平均错误率,寻找Wisconsin Diagnostic Breast Cancer数据集中可以保证敏感度和准确度的属性约简。通过实验分析,新的方法可以获得更高的精度,并发现乳腺癌的诊断与质地、光滑度和周长密切相关。

An improved method based on the lower approximation distribution identical is proposed. However the accuracy of individual decision classes is low so it is not conducive to the diagnosis of breast cancer. According to this problem, the methods is used based on lower approximation distribution. The presence probability and average error rate of each attribute are based on to look for attribute reduction that guarantees sensitivity and accuracy in the Wisconsin Diagnostic Breast Cancer dataset. Through experimental analysis, new method can achieve higher accuracy,and it has found that diagnosis of breast cancer is closely related to texture, smoothness and perimeter.