基于得分函数的概率分类模型研究

Research on Probability Classification Model Based on Scoring Function

摘要: 现代统计学中有各种分类方法，在数据研究中，类别分得越精准，得到的结果就越有价值。对于二元分类问题，本文提出了一种基于得分函数的概率分类模型MKL，从理论上证明了所提出的MKL估计的一致性。在实证方面，本文通过拟牛顿算法直接对连续化后的MKL统计量进行优化，给出了模拟研究的分类效果和一个心脏衰竭数据集的实例。该方法考虑了预测能力、计算复杂度和实际可解释性方面的权衡，与现有的分类方法相比具有优势。

Abstract: There are various classification methods in modern statistics, and in data research, the more accu-rate the classification, the more valuable the results obtained. For the binary classification problem, this paper proposes a probabilistic classification model MKL based on the score function, which theoretically proves the consistency of the proposed MKL estimate. In terms of empirical evidence, this paper directly optimizes the continuous MKL statistics by means of quasi-Newtonian algorithm, and gives the classification effect of the simulation study and an example of a heart failure dataset. This approach takes into account trade-offs in terms of predictive power, computational complexity, and practical interpretability, and offers advantages over existing classification methods.