基于LAZE先验的机器学习方法在癌症数据上的应用——以前列腺癌数据集为例

针对高维癌症数据的特征构建稀疏化模型,分析了已有的稀疏化方法能够取得稀疏化结果的原因,以此为基础改进了使用LAZE先验的贝叶斯方法,得到了2个适用于癌症数据的新稀疏化方法（使用半混合先验的贝叶斯方法和使用完整LAZE先验的贝叶斯方法）;并利用前列腺癌数据集对2个新稀疏化方法的可行性与有效性进行验证. 数值实验结果表明:与传统稀疏化方法相比较,新稀疏化方法不仅能够得到较好的稀疏效果,能够完全剔除与目标指标无关的临床指标,还能得到较低的误差.

Application of Machine Learning Methods Based on LAZE Priors to Cancer Data——Take the Prostate Cancer Data Set for Example

A sparse model according to the characteristics of high-dimensional cancer data is constructed, the reasons why existing sparse methods can achieve sparse results is analyzed, and based on this, the Bayesian method using LAZE prior is improved, and two new sparse methods suitable for cancer data are obtained: Bayesian method using semi-mixed prior and Bayesian method using complete LAZE prior. The feasibility and effectiveness of the two new sparse methods obtained in this study are verified by using prostate cancer dataset. Numerical experimental results show that compared with traditional sparse methods, the new sparse methods can not only get better sparse effect and completely remove clinical indicators unrelated to the target indicators but also decrease errors.