L1+L2正则化逻辑斯蒂模型分类算法

Logistic Model Classification Algorithm via L1+L2 Regularization

提出一种L1+L2范数正则化逻辑斯蒂模型分类算法。该算法引入L2范数正则化,解决L1正则化逻辑斯蒂算法迭代过程奇异问题,通过引入样本向量的扩展和新的权值向量完成L1范数非平滑问题,最终使用共轭梯度方法求解经过转化的最优化问题。在各种实际数据集上的实验结果表明,该算法优于L2范数、L1范数和Lp范数正则化逻辑斯蒂模型,具有较好的特征选择和分类性能。

This paper proposes an L1＋L2 norm regularized logistic model classification algorithm, and the singularity of iterative process in L1 norm regularized logistic classification algorithm is solved by using L2 norm regularization. The non-smooth problem is transformed into smooth one via argumentation of vector of samples and introduction of new weight vector, and classification object function is solved using the conjugate gradient method. Performance of classification and feature selection on real datasets shows that the algorithm is better than L2 norm, L1 nrom and Lp norm regularized logistic model.