Semi-Supervised Classifier Ensemble with High-Quality Subspace for High-Dimensional Data
Guojie Li, Kaixiang Yang, Zhiwen Yu · 2023
Due to the noisy and high-dimensional character of the data combined with a small sample size, semi-supervised classification of high-dimensional data with few labeled samples is a significant problem in machine learning and data mining. To address these challenges, we first propose a novel feature evaluation function that combines the advantages of the Laplacian score and intra-class and inter-class smoothness score to select high-quality features. Subsequently, we combine the selected high-quality features with some randomly selected features to generate a set of high-quality subspaces (HQS). Building upon the concept of HQS, we further propose an ensemble framework (E-HQS) specifically designed for semi-supervised classification tasks on high-dimensional data. We thoroughly validate the effectiveness of our proposed method on 8 real high-dimensional datasets.