Exploração das propriedades de hubness para detecção semissupervisionada de outliers em dados de alta dimensão

Abstract

With the increase in the amount of data stored, the area of data mining has become essential for it to be possible to manipulate and extract knowledge from these data. Much of the work in this area focuses on finding patterns in the data, but non-standard data (anomalies) can also add much to the knowledge of the data set under study. The study, development and enhancement of outliers detection techniques are important objectives and have proven useful in several scenarios, such as: fraud detection, intrusion detection and monitoring of medical conditions, among others. The paper presented here describes a novel method for semi-supervised detection of outliers in high dimensional data. Experiments with several real datasets indicate the superiority of the proposed method in relation to the literature methods selected as the baseline.