Every task new to the user can be approached advantageously with f(x) = y.

In `R`

terms f, x \& y are *objects*.

x is the object at hand, y is the object desired and f is the object, acting as a function, that transforms x to y.

I'm going to speculate about x, and adjustments to the approach outlined will be required based on the actual data.

Let x be organized as a `tidy`

data frame keyed to the date and time of observation as a column (Time is needed only for the purpose of ensuring unique rows).

The form of a predictive model is Y ~ X_i...X_n + \epsilon

Y is the response variable, which will have a great influence on the choice of f. It may be continuous, categorical or binary, depending on how the data was collected and the characteristic of interest.

A Y about population characteristics might be a census count, a binning into categories (low, medium, high) or presence/absence. For some species, such as ants, the continuous count might be meaningful, for others that are encountered in smaller quantity, it may be the second and for family unit species, binary.

The X covariates are categorical, which may be similarly classified. Categories of stream density is one example.

P(Y|X_i) can be tested with many possible f. The first step, however, is properly to understand the nature of x.