Abstract Machine learning based outlier detection methods are widely used in various domains. However, an ensemble of such detection methods could leverage detection performance. The existing ensemble methods made up of multiple unsupervised learning algorithms lack in ideal strategy for choosing right candidates as constituent detectors. It resulted in mediocrity in model stability and accuracy. To overcome this problem, in this paper, we propose an ensemble framework based outlier detection system in high dimensional data. It has ideal mechanism for effectively choosing base outlier detectors. Out of many candidate outlier detectors, the ones that yield highest performance are combined. An algorithm named Average Selection and Ensemble of Candidates for Outlier Detection (ASEC-OD). Many real world datasets are used for empirical study. The results of experiments revealed that the proposed framework outperforms many existing methods.