Traitement des valeurs aberrantes : concepts actuels et tendances générales

Treatment of outliers: present state of concepts and general tendencies. The fast evolution of data collection means and the computerised treatment of information induced a huge problem of outliers during last decades. For example, development of geographic information systems, precision farming and automatisation of data capture led to the constitution of large databases where outliers can more easily go unnoticed. The presence of abnormal values may lead to biased estimations of parameters of populations and, further to the realization of statistical tests to interpretation of results, which can be erroneous. This article highlights the diversity of available methods and emphasizes the way of handling outliers in a structured way. In spite of very widely developed theoretical concepts and a large bibliography on the subject, we notice that most of existing statistical software packages on the market are very limited for the treatment of outliers.