In Piedmont (NW Italy) the environmental changes due to human impact have deeply altered the white-clawed crayfish habitat and native populations have decreased markedly. The evaluation of the chemical-physical factors determining its presence can contribute to its conservation. The study system consisted of 175 sites and Austropotamobius pallipes was recorded in 98. Two different approaches were used to assess its presence: Logistic Regression (LR) and Decision Tree (DT) models. The data were normalized proportionally (between 0-1) before a data set was used to build different models. Attributes were selected through the Information Gain method and a subset of 9 inputs (NH4+, NO3-, PO43-, SO42-, Ca2+, Mg2+, water hardness, pH, SpO2) was obtained from the starting one of 12. The performances of LR were estimated from a leave-one-out jack knifing involving a holdout procedure repeated 10 times using a model derived from a calibration set of 80% of the sites. In building DT models, the J48 algorithm was used with a binary split and the tree-pruning optimization method was applied. DT training and validation were based on stratified 10-fold cross-validation and to estimate a reliable error of the models, experiments were repeated 10 times. The percentage of correctly classified instances was: LR = 67.30% (sensitivity = 82.81%, specificity = 47.21%), DT = 63.76% (sensitivity = 75.06%, specificity = 49.71%). The Mann-Whitney U test showed that the LR performed better than the DT. The concentration of Ca2+, water hardness, and BOD5 were the most important inputs used in building LR models.
Comparison among different approaches to model chemical-physical water requirements of white-clawed crayfish (Austropotamobius pallipes) in North Western Italy
FAVARO, LIVIO;TIRELLI, Santina;PESSANI, Daniela
2010-01-01
Abstract
In Piedmont (NW Italy) the environmental changes due to human impact have deeply altered the white-clawed crayfish habitat and native populations have decreased markedly. The evaluation of the chemical-physical factors determining its presence can contribute to its conservation. The study system consisted of 175 sites and Austropotamobius pallipes was recorded in 98. Two different approaches were used to assess its presence: Logistic Regression (LR) and Decision Tree (DT) models. The data were normalized proportionally (between 0-1) before a data set was used to build different models. Attributes were selected through the Information Gain method and a subset of 9 inputs (NH4+, NO3-, PO43-, SO42-, Ca2+, Mg2+, water hardness, pH, SpO2) was obtained from the starting one of 12. The performances of LR were estimated from a leave-one-out jack knifing involving a holdout procedure repeated 10 times using a model derived from a calibration set of 80% of the sites. In building DT models, the J48 algorithm was used with a binary split and the tree-pruning optimization method was applied. DT training and validation were based on stratified 10-fold cross-validation and to estimate a reliable error of the models, experiments were repeated 10 times. The percentage of correctly classified instances was: LR = 67.30% (sensitivity = 82.81%, specificity = 47.21%), DT = 63.76% (sensitivity = 75.06%, specificity = 49.71%). The Mann-Whitney U test showed that the LR performed better than the DT. The concentration of Ca2+, water hardness, and BOD5 were the most important inputs used in building LR models.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
