Prioritizing removals of a highly cryptic invasive species using machine learning and demographic weighting

This repository contains all data, R code, fitted model objects, machine‑learning outputs, and figures that support the unpublished manuscript “Prioritizing removals of a highly cryptic invasive species using machine learning and demographic weighting”. The study introduces the Weighted Removal Index (WRI), a demographic weighting scheme that scales each captured Burmese python by its estimated survivorship and reproductive value and evaluates three modelling frameworks (GAM, BRT and NN) for predicting both capture success and survey WRI. All materials are released under CC‑BY 4.0. Cite this Zenodo record when using the data or code. File Type Purpose / key contents 240819_PEP_Demo.R R script Builds the Weighted Removal Index (WRI), imputes missing SVLs, engineers survey-level predictors, fits and evaluates the GAM and BRT models, analyzes contractor performance, and compares model predictions on a validation set. 240819_PEP_Demo_Machine Learning.R R script TensorFlow/Keras workflow: builds balanced‑accuracy‑optimised capture classifier; appends scaled capture probabilities; tunes Tweedie‑loss WRI regressor; performs Bayesian hyper‑parameter search; saves predictions PEP_manuscript.csv CSV Screened survey records (n = 4 092) used in the manuscript; one row = one survey; includes linked snake IDs, environmental & operational covariates. PEP_captures.csv CSV Contractor capture database v4.0; sex, total & snout‑vent length, weight, geolocation, etc. PEP3_0_Flattened_Simplified_052720_020521_EFM.csv CSV Legacy capture database v3.0 used to back‑fill demographics for older snake IDs. PEP_wri.rds RDS Fully processed survey‑level dataset with all engineered predictors and the two WRI measures (wri, wri.surv) ready for modelling. Generated at the end of 240819_PEP_Demo.R. brt_gridsearch.rds RDS List produced by gbm3::grid_search(): every combination of trees, depth, shrinkage & min‑obs tested plus CV metrics; best parameters used in the manuscript BRT. gam_stepwiseAIC.rds RDS Data‑frame with one row per family‑link candidate (n = 30) and columns: model (character call), deviance explained DE, error flag, and any warning notes. nn_pred.csv CSV Hold‑out‐test predictions from the final neural‑network WRI model; columns: survey ID, observed WRI, predicted WRI, absolute & squared error.