An Ensemble Outlier Detection Method Based on Information Entropy-Weighted Subspaces for High-Dimensional Data

Outlier detection is an important task in the field of data mining and a highly active area research machine learning. In industrial automation, datasets are often high-dimensional, meaning effort to study all dimensions directly leads sparsity, thus causing outliers be masked by noise effects high-dimensional spaces. The “curse dimensionality” phenomenon renders many conventional outlier methods ineffective. This paper proposes new algorithm called EOEH (Ensemble Detection Method Based on Information Entropy-Weighted Subspaces for High-Dimensional Data). First, random secondary subsampling performed data, detectors run various small-scale sub-samples provide diverse results. Results then aggregated reduce global variance enhance robustness algorithm. Subsequently, information entropy utilized construct dimension-space weighting method that can discern influential factors within different dimensional generates weighted subspaces objects, reducing impact created improving performance. Finally, this offers design high-precision local factor (HPLOF) detector amplifies differentiation between normal thereby performance feasibility validated through experiments used both simulated UCI datasets. comparison popular algorithms, our demonstrates superior runtime efficiency. Compared with current popular, common improves 6% average. terms running time 20% faster than algorithms.