Anomaly detection is an important issue in many application domains. For example, cybercrime can cause considerable economic losses and threaten companies survival. Securing its information system has become a priority and a strategic issue for all types of companies. Other areas are also be impacted such as health, transport, etc. The implemented supervision solutions are often based on anomaly detection algorithms from datamining and machine learning domains. We present in this paper a complete state of the art on anomaly detection algorithms. We propose a classification of these methods based on the type of data sets (flows, time series, graphs, etc.), the application domain and the considered approach (statistics, classification, clustering, etc.). We then focus on three algorithms: LOF, OC-SVM and Isolation Forest, that we test on two different datasets to compare their performance.