Is On-Line Data Analysis Safety? Pitfalls Steaming from Automated Processing of Heterogeneous Environmental Data and Possible Solutions
| Authors | |
|---|---|
| Year of publication | 2011 |
| Type | Article in Proceedings |
| Conference | Environmental Software Systems: Frameworks of Environment, IFIP Advances in Information and Communication Technology, vol. 359 |
| MU Faculty or unit | |
| Citation | |
| Field | Other medical specializations |
| Keywords | classification; nonparametric multivariate analysis; heterogeneous data |
| Description | The current situation in environmental monitoring is characterized by increasing amount of data from monitoring networks together with increasing requirements on joining of these data from various sources in comprehensive databases and their usage for decision support in environmental protection and management. The automated analysis of such a heterogeneous datasets is a complicated process, rich in statistical pitfalls. There is a number of methods for multivariate classification of objects, e.g. logistic regression, discriminant analysis or neural networks; however, most of commonly used classification techniques have prerequisites about distribution of data, are computationally demanding or their model can be considered as “black box”. Keeping these facts in mind, we attempted to develop a robust multivariate method suitable for classification of unknown cases with minimum sensitivity to data distribution problems; and thus, suitable for routine use in practice. |