Data-based decision-making requires correct statistical inference and modelling. Real data often contain points that significantly differ from other observations. These points could be
a) registration or measurement errors or
b) true data values that are extremal or atypical.
The latter appear especially in the case of natural disasters, management anomalies, financial crises and other extremal events or superposition of multiple processes. In the first case, their detection is useful for data cleaning, while in the second case, they can uncover interesting or unusual properties of the observed random variable that may have gone unnoticed. With the growth of the data size, the risk of observing such examples increases significantly. The most dangerous situations are related to the cases when, both a) and b), appear as outside values or extremes in the sample. Methods for their analysis depend on the structure of the dataset and the aim of the analysis.
Along the project, we are going to developed algorithms for anomaly detection in ecology and management.