Species 061: Juniperus ashei

General Linear Models

Boosted Regression Trees

Generalized Linear Model Results

Data:

C:\Withrow\Workspace\Model\glm_1\CovariateCorrelationOutputMDS_initial.csv

n(pres)                 =      576

n(abs)                  =      5633

n covariates considered =      4

total time for model fitting = 0.27min

Settings:

model family          : binomial

simplification method : AIC

Results:

number covariates in final model   : 9

Call:

glm(formula = response ~ bio01 + I(bio01^2) + bio03 + I(bio03^2) + 

    I(bio15^2) + bio15 + bio08 + bio01:bio15 + bio03:bio08, family = model.family, 

    data = dat, weights = weight, na.action = "na.exclude")

Deviance Residuals: 

    Min       1Q   Median       3Q      Max  

-2.2982  -0.0713  -0.0017   0.0000   4.5357  

Coefficients:

              Estimate Std. Error z value Pr(>|z|)    

(Intercept) -3.534e+02  2.704e+01 -13.071  < 2e-16 ***

bio01        9.573e-01  1.287e-01   7.440 1.00e-13 ***

I(bio01^2)  -3.446e-03  3.629e-04  -9.496  < 2e-16 ***

bio03        1.035e+01  1.086e+00   9.531  < 2e-16 ***

I(bio03^2)  -7.558e-02  1.268e-02  -5.962 2.49e-09 ***

I(bio15^2)  -5.786e-03  7.162e-04  -8.079 6.52e-16 ***

bio15       -9.318e-01  1.676e-01  -5.559 2.72e-08 ***

bio08        6.491e-01  1.247e-01   5.206 1.92e-07 ***

bio01:bio15  7.154e-03  8.180e-04   8.745  < 2e-16 ***

bio03:bio08 -1.664e-02  3.183e-03  -5.229 1.71e-07 ***

---

Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 3835.9  on 6208  degrees of freedom

Residual deviance: 1512.3  on 6199  degrees of freedom

AIC: 1532.3

Number of Fisher Scoring iterations: 11

============================================================ 

Evaluation Statistics applied to train split:

Correlation Coefficient      : 0.7600114 

NULL Deviance                : 0.61779

Fit Deviance                 : 0.24357

Explained Deviance           : 0.37422

Percent Deviance Explained   : 60.574

  Threshold Methods based on Sens=Spec 

Threshold                    :  0.18 

Confusion Matrix: 

         observed

predicted    1    0

        1  533  419

        0   43 5214

AUC                          :  0.9725 

Percent Correctly Classified :  92.55919 

Sensitivity                  :  0.9253472 

Specificity                  :  0.9256169 

Kappa                        :  0.6581244 

True Skill Statistic         :  0.8509641 

   Calibration Statistics 

Intercept (general calibration)                            :  0.001056991 

Slope   (direction and variation in fit)                   :  1.002383 

Testa0b1 (overall reliability of predictors)               :  0.9985342 

Testa0|b1(incorrect calibration given correct refinement)  :  0.9967887 

Testb1|a (refinement given correct calibration)            :  0.9569237 

============================================================ 

Evaluation Statistics applied to crossValidation split:

Correlation Coefficient      : 0.7573343  (sd 0.026456) 

NULL Deviance                : 0.61779 (sd 0.0034313)

Fit Deviance                 : 0.24962 (sd 0.028979)

Explained Deviance           : 0.36817 (sd 0.029355)

Percent Deviance Explained   : 59.592 (sd 4.7056)

  Threshold Methods based on Sens=Spec 

Mean Threshold               : 0.172 (sd 0.011353) 

Confusion Matrix: 

         observed

predicted    1    0

        1  530  429

        0   46 5204

AUC                          :  0.97139  (sd 0.0059164) 

Percent Correctly Classified :  92.34948  (sd 0.93347) 

Sensitivity                  :  0.9201149  (sd 0.012345) 

Specificity                  :  0.9238414  (sd 0.010535) 

Kappa                        :  0.6509013  (sd 0.031839) 

True Skill Statistic         :  0.8439563  (sd 0.014097) 

   Calibration Statistics 

Intercept (general calibration)                            :  0.0162156  (sd 0.20833) 

Slope   (direction and variation in fit)                   :  1.010856  (sd 0.16993) 

Testa0b1 (overall reliability of predictors)               :  0.4666593  (sd 0.24584) 

Testa0|b1(incorrect calibration given correct refinement)  :  0.5912218  (sd 0.2644) 

Testb1|a (refinement given correct calibration)            :  0.3432081  (sd 0.20215) 

Total time = 3.81 min

Boosted Regression Tree Modeling Results

Data:

C:\Withrow\Workspace\Model\brt_1\CovariateCorrelationOutputMDS_initial.csv

n(pres)                 =      576

n(abs)                  =      5633

n covariates considered =      4

total time for model fitting = 0.28min

Settings:

random seed used             : 19680828

tree complexity              : 10

learning rate                : 0.0699

n(trees)                     : 300

model simplification         : cross-validation

n folds                      : 3

n covariates in final model  : 3

Relative influence of predictors in final model:

   Var  rel.inf

 bio01 45.44683

 bio08 28.35484

 bio15 26.19834

Important interactions in final model:

 v1 name1 v2 name2

  2 bio08  1 bio01

  3 bio15  1 bio01

============================================================ 

Evaluation Statistics applied to train split:

Correlation Coefficient      : 0.8679842 

NULL Deviance                : 0.61779

Fit Deviance                 : 0.14172

Explained Deviance           : 0.47608

Percent Deviance Explained   : 77.061

  Threshold Methods based on Sens=Spec 

Threshold                    :  0.2 

Confusion Matrix: 

         observed

predicted    1    0

        1  548  260

        0   28 5373

AUC                          :  0.9916 

Percent Correctly Classified :  95.36157 

Sensitivity                  :  0.9513889 

Specificity                  :  0.9538434 

Kappa                        :  0.7666289 

True Skill Statistic         :  0.9052323 

   Calibration Statistics 

Intercept (general calibration)                            :  0.2407863 

Slope   (direction and variation in fit)                   :  1.445837 

Testa0b1 (overall reliability of predictors)               :  2.561285e-13 

Testa0|b1(incorrect calibration given correct refinement)  :  0.9435936 

Testb1|a (refinement given correct calibration)            :  2.642331e-14 

============================================================ 

Evaluation Statistics applied to crossValidation split:

Correlation Coefficient      : 0.7432669  (sd 0.066827) 

NULL Deviance                : 0.61779 (sd 0.0034313)

Fit Deviance                 : 0.24885 (sd 0.045032)

Explained Deviance           : 0.36894 (sd 0.042518)

Percent Deviance Explained   : 59.745 (sd 7.1286)

  Threshold Methods based on Sens=Spec 

Mean Threshold               : 0.204 (sd 0.0069921) 

Confusion Matrix: 

         observed

predicted    1    0

        1  490  331

        0   86 5302

AUC                          :  0.96924  (sd 0.011122) 

Percent Correctly Classified :  93.28439  (sd 1.5028) 

Sensitivity                  :  0.8508772  (sd 0.049797) 

Specificity                  :  0.9412403  (sd 0.013984) 

Kappa                        :  0.6669422  (sd 0.061914) 

True Skill Statistic         :  0.7921174  (sd 0.056387) 

   Calibration Statistics 

Intercept (general calibration)                            :  -0.07622476  (sd 0.2577) 

Slope   (direction and variation in fit)                   :  0.8740694  (sd 0.11683) 

Testa0b1 (overall reliability of predictors)               :  0.2570407  (sd 0.26775) 

Testa0|b1(incorrect calibration given correct refinement)  :  0.5054031  (sd 0.2897) 

Testb1|a (refinement given correct calibration)            :  0.1994819  (sd 0.23524) 

Total time = 3.36 min

MARS

Random Forests

MARS Model Results

Data:

C:\Withrow\Workspace\Model\mars_1\CovariateCorrelationOutputMDS_initial.csv

n(pres)                 =      576

n(abs)                  =      5633

n covariates considered =      4

total time for model fitting = 0.05min

Settings:

random seed used             : 123

mars degree                  : 1

mars penalty                 : 2

Summary of Model:

      nsubsets   gcv    rss

bio01       13 100.0  100.0

bio03       11  58.9   59.3

bio15        8  44.7   45.1

bio08        6  31.8   32.4

============================================================ 

Evaluation Statistics applied to train split:

Correlation Coefficient      : 0.7714834 

NULL Deviance                : 0.61779

Fit Deviance                 : 0.234

Explained Deviance           : 0.38379

Percent Deviance Explained   : 62.123

  Threshold Methods based on Sens=Spec 

Threshold                    :  0.16 

Confusion Matrix: 

         observed

predicted    1    0

        1  528  439

        0   48 5194

AUC                          :  0.9721 

Percent Correctly Classified :  92.15655 

Sensitivity                  :  0.9166667 

Specificity                  :  0.9220664 

Kappa                        :  0.6428534 

True Skill Statistic         :  0.8387331 

   Calibration Statistics 

Intercept (general calibration)                            :  0.0001724258 

Slope   (direction and variation in fit)                   :  1.000616 

Testa0b1 (overall reliability of predictors)               :  0.9998777 

Testa0|b1(incorrect calibration given correct refinement)  :  0.9966712 

Testb1|a (refinement given correct calibration)            :  0.9879749 

============================================================ 

Evaluation Statistics applied to crossValidation split:

Correlation Coefficient      : 0.7696257  (sd 0.02646) 

NULL Deviance                : 0.61779 (sd 0.0034313)

Fit Deviance                 : 0.23808 (sd 0.023503)

Explained Deviance           : 0.37971 (sd 0.023416)

Percent Deviance Explained   : 61.464 (sd 3.7925)

  Threshold Methods based on Sens=Spec 

Mean Threshold               : 0.158 (sd 0.0042164) 

Confusion Matrix: 

         observed

predicted    1    0

        1  527  449

        0   49 5184

AUC                          :  0.97127  (sd 0.0061404) 

Percent Correctly Classified :  91.97914  (sd 1.2672) 

Sensitivity                  :  0.914882  (sd 0.015577) 

Specificity                  :  0.9202921  (sd 0.014449) 

Kappa                        :  0.6384392  (sd 0.040887) 

True Skill Statistic         :  0.8351742  (sd 0.017014) 

   Calibration Statistics 

Intercept (general calibration)                            :  0.0418931  (sd 0.31402) 

Slope   (direction and variation in fit)                   :  1.023892  (sd 0.1745) 

Testa0b1 (overall reliability of predictors)               :  0.4354009  (sd 0.30875) 

Testa0|b1(incorrect calibration given correct refinement)  :  0.558643  (sd 0.32263) 

Testb1|a (refinement given correct calibration)            :  0.3265917  (sd 0.25268) 

Total time = 0.68 min

Random Forest Modeling Results

Data:

C:\Withrow\Workspace\Model\rf_1\CovariateCorrelationOutputMDS_initial.csv

n(pres)                 =      576

n(abs)                  =      5633

n covariates considered =      4

total time for model fitting = 0.11min

Settings:

random seed used                       : 19680828

n covariates considered at each split  : 2

n trees                                : 1000

Relative performance of predictors in final model:

            0        1 MeanDecreaseAccuracy MeanDecreaseGini

bio01 53.8396  71.8018             110.0564         221.9335

bio03 46.5026 104.9509              81.3087         115.5304

bio08 45.9911  39.1408              65.5120         125.5323

bio15 43.5092  92.6888              65.3771         170.7886

============================================================ 

Evaluation Statistics applied to train split:

Correlation Coefficient      : 0.8135087 

NULL Deviance                : 0.61779

Fit Deviance                 : 0.20033

Explained Deviance           : 0.41747

Percent Deviance Explained   : 67.574

  Threshold Methods based on Sens=Spec 

Threshold                    :  0.09 

Confusion Matrix: 

         observed

predicted    1    0

        1  539  363

        0   37 5270

AUC                          :  0.9799 

Percent Correctly Classified :  93.55774 

Sensitivity                  :  0.9357639 

Specificity                  :  0.9355583 

Kappa                        :  0.6948069 

True Skill Statistic         :  0.8713222 

   Calibration Statistics 

Intercept (general calibration)                            :  -0.147462 

Slope   (direction and variation in fit)                   :  0.8779092 

Testa0b1 (overall reliability of predictors)               :  0.000727222 

Testa0|b1(incorrect calibration given correct refinement)  :  0.5563412 

Testb1|a (refinement given correct calibration)            :  0.0001727478 

============================================================ 

Evaluation Statistics applied to crossValidation split:

Correlation Coefficient      : 0.8132128  (sd 0.022876) 

NULL Deviance                : 0.61779 (sd 0.0034313)

Fit Deviance                 : 0.19798 (sd 0.022205)

Explained Deviance           : 0.4198 (sd 0.020559)

Percent Deviance Explained   : 67.961 (sd 3.4964)

  Threshold Methods based on Sens=Spec 

Mean Threshold               : 0.3365 (sd 0.042035) 

Confusion Matrix: 

         observed

predicted    1    0

        1  481  173

        0   95 5460

AUC                          :  0.97858  (sd 0.0055847) 

Percent Correctly Classified :  95.68373  (sd 0.38517) 

Sensitivity                  :  0.8350877  (sd 0.037519) 

Specificity                  :  0.9692891  (sd 0.0057945) 

Kappa                        :  0.758361  (sd 0.018529) 

True Skill Statistic         :  0.8043768  (sd 0.033773) 

   Calibration Statistics 

Intercept (general calibration)                            :  -0.1117316  (sd 0.25667) 

Slope   (direction and variation in fit)                   :  0.8591193  (sd 0.088338) 

Testa0b1 (overall reliability of predictors)               :  0.3341178  (sd 0.28532) 

Testa0|b1(incorrect calibration given correct refinement)  :  0.5970467  (sd 0.35561) 

Testb1|a (refinement given correct calibration)            :  0.2384026  (sd 0.22122) 

Total time = 2 min