A Hybrid Unsupervised Density-based Approach with Mutual Information for Text Outlier Detection

The detection of outliers in text documents is a highly challenging task, primarily due to the unstructured nature and curse dimensionality. Text document refer data that deviates from found other belonging same category. Mining has wide applications various domains, including spam email identification, digital libraries, medical archives, enhancing performance web search engines, cleaning corpora used classification. To address issue dimensionality, it crucial employ feature selection techniques reduce large number features without compromising their representativeness domain. In this paper, we propose hybrid density-based approach incorporates mutual information for outlier detection. proposed utilizes normalized identify most distinct characterize target Subsequently, customize well-known local factor algorithm suit datasets. evaluate effectiveness approach, conduct experiments on synthetic real datasets comprising twelve high-dimensional results demonstrate consistently outperforms conventional methods, achieving an average improvement 5.73% terms AUC metric. These findings highlight remarkable enhancements achieved by leveraging conjunction with algorithm, particularly