Outlier Detection using Ensemble Learning
Moirangthem Marjit Singh, Nongmaithem Kane · 2022
The growth of internet in recent years have led to emergence of several types of network attacks and outlier. Hence development of techniques to deal with network attacks and outlier has become an important area of research. Several methods for detecting outliers have been found in the current literature. In this paper, we propose a method for outlier detection which is based on ensemble learning and other well-known methods like LOF, KNN, HBOS, iForest, COPOD and PCA as the component of our ensemble model for base estimator. The proposed method deals with highdimensional data by projecting the original data (JL projection) onto a sub space with smaller number of samples and for each data point their anomaly score will be calculated. There after the scores are combined using some common combination methods i.e., score aggregation, maximization and majority voting. The base estimator of our ensemble model as well as other existing models are trained and evaluated using seven benchmark datasets. The proposed model delivered higher detection performance in terms of accuracy and ROC. The use of ensemble learning approach is found to be helpful in dealing with high dimensional data and unsupervised methods with respect to outlier detection.