Supervised outlier detection for classification and regression

Outlier detection, i.e., the task of detecting points that are markedly different from data sample, is an important challenge in machine learning. When a model built, these special can skew training and result less accurate predictions. Due to this fact, it identify remove them before building any supervised often first step when dealing with learning problem. Nowadays, there exists very large number outlier detector algorithms provide good results, but their main drawbacks unsupervised nature together hyperparameters must be properly set for obtaining performance. In work, new estimator proposed. This done by pipelining following model, such way targets later supervise how all involved optimally selected. pipeline-based approach makes easy combine detectors classifiers regressors. experiments done, nine relevant have been combined three regressors over eight regression problems as well two another binary multi-class classification problems. The usefulness proposal objective automatic determine has proven effectiveness also analyzed compared.