基于张量CP分解的不完整数据自适应图特征选择OA

The feature selection problem for tensor data is a complex task that requires comprehensive consider-ation of the multi-dimensional structure information of the data and incomplete data.To solve this problem,an adap-tive graph regularization feature selection algorithm for incomplete data based on tensor CP(Canonical Polyadic)de-composition is proposed.The collaborative design of data filling and feature selection reduces the high computational load of the traditional staged processing method,effectively improving the learning efficiency.The tensor CP decom-position is utilized to precisely fill the incompleteness of the data.The introduction of an adaptive graph structure based on the Gini coefficient can more accurately embed the geometric structure of the data into manifold learning,which not only effectively enhances the robustness and computational efficiency of the model,but also obtains the op-timal feature subset of the data.Comparison results of six feature selection algorithms on six public datasets verifies the effectiveness of the proposed algorithm.