Menu boczne

Treść strony

Ann Acad Med Stetin, 2004; 50, Suppl. 1, 47-53

IZABELA GUTOWSKA, ZYGMUNT MACHOY, DARIUSZ CHLUBEK, BOGUSŁAW MACHALIŃSKI*

LIVING CONDITIONS OF DEER IN THE PROVINCES OF WESTERN POMERANIA
AND LUBUSKIE AS REVEALED BY MANDIBULAR CONTENT OF FLUORIDE,
CALCIUM AND MAGNESIUM.
II. DETAILED STATISTICAL ANALYSIS WITH THE STATISTICA NEURAL
NETWORKS SOFTWARE

Katedra i Zakład Biochemii i Chemii Pomorskiej Akademii Medycznej w Szczecinie
al. Powstańców Wlkp. 72, 70-111 Szczecin
Kierownik: prof. dr hab. n. med. Dariusz Chlubek
* Zakład Patologii Ogólnej Pomorskiej Akademii Medycznej w Szczecinie
al. Powstańców Wlkp. 72, 70-111 Szczecin
Kierownik: dr hab. n. med. Bogusław Machaliński

Summary
Animals from areas contaminated by industrial emissions containing fluoride may accumulate it predominantly in hard tissues. The stable composition of the mandible makes it suitable for the study of accumulation of fluorides and other elements (Mg, Ca). It is also possible to determine the extent of pollution with fluorine compounds around industrial plants and to reveal some of the features of the natural habitat of wild animals. We applied the advanced technique of neural networks to investigate the simultaneous influence of multiple parameters on the accumulation of fluorine, magnesium, and calcium in the mandible of deer and possible interactions between these elements. The first step involved the generation of a neural network. In the present work, networks were created with the Intelligent Problem Solver (IPS). Input data were divided into three sets. One was used for training (Tr) of the network, another for validation (Ve) and control over the training process and the last one for testing (Te). Network quality was checked with the following parameters: Data Mean – average value of variable; Data SD – standard deviation of variable; Error Mean – average value of error; Abs. Mean – average error modulus; Error SD – standard deviation of error; SD Ratio – ratio of standard deviation of error to standard deviation of variable (Error SD/Data SD); Correlation – Pearson’s correlation coefficient. For the model to be satisfactory, Pearson’s correlation coefficient should be close to 1. In the case of complex problems with an unclear influence of parameters on the process, an analysis of parametric sensitivity must be performed. This task was accomplished automatically with the computer program. It was found that the effect of the same parameters on the accumulation of fluorine, magnesium and calcium varied. The results suggest interactions between the examined elements.

K e y w o r d s: statistical analysis – artificial neural network – fluorine – calcium – magnesium.

Powrót
do góry