混合数据下基于Tsallis熵的集成因果特征选择算法OA

With the development of intelligent technologies,data has become high-dimensional.Causal feature selection,which is highly interpretable and robust,has been widely studied.It aims to select the feature set most relevant to the class variable,namely its parents and children(PC)and spouses(SP)sets,collectively referred to as the Markov blanket(MB)set,to reveal the causal relationships between features.In practical application scenarios,data often exhibits mixed distri-butions.Most existing causal feature selection algorithms focus only on single-type data and cannot be directly applied to mixed data.To address the above issue,this paper proposes an ensemble causal feature selection algorithm based on Tsallis entropy,called HTECFS,in mixed data scenarios.During the PC learning process,Tsallis mutual information,calculated directly from different forms of Tsallis entropy,is used to measure the correlation between features,avoiding potential information loss caused by data preprocessing.During the SP learning process,Tsallis conditional mutual information is used to test the SP features on the PC,and newly learned SP features are incorporated into the MB set to improve predic-tion accuracy of the model.Furthermore,multiple learning subspaces are constructed for independent MB learning to obtain an approximately globally optimal MB set.Moreover,a group optimization strategy is introduced to fuse the learning results,enhancing the globality and stability of the feature subset.Extensive tests on ten different types of datasets fully validate the effectiveness of the proposed algorithm.