cv_wells.csv is a tabular data frame with all the attribute data necessary to run ML models to predict the dry column. Plot shp/cv_wells.shp if you want to visualize the locations of the data and the spatial dependence of well failure.
In the csv, the first column, name, is the unique ID of the well. There are no duplicate well entries.
The second column, dry is the binary response variable. 1 indicates a dry well, and 0 indicates a non-dry well.
The remaining columns are geologic and bioclimatic variables from the USGS and WorldClim extracted from rasters to the spatial points.
That leaves, for the model specification:
dry ~ hyd_cond + tmin1...tmin12 + tmax1...tmax12 + prec1...prec12 + bio1...bio12
IMPORTANT: the ratio of failures to non-failures, or 1/0, in the dataset is around 5%. This means that it's possible to achieve a training accuracy of 95% with a null model by simply assuming that no observation fails. We've navigated around this by equally sampling the failure and non-failure sets.
In the future, we should compare the results from multiple non-failure sets as a way to bootstrap different non-failure sets and provide an estimate of the uncertainty inherent in our random selection of non-failing points.