MLmorph

MLmorph integrates morphological modeling with machine learning (random forests) to support structured decision-making. The package enumerates a morphospace—the set of feasible configurations across selected variables—and estimates class probabilities over that space. It includes:

• utilities for factorizing inputs into analysis-ready factors,
• a wrapper to train random forest classifiers and report validation performance,
• construction of morphospaces with predicted probabilities and a flag for purely simulated combinations,
• a Shiny app for interactive scenario exploration.

Installation

```r

install.packages("devtools")

devtools::install_github("theogrost/MLmorph") ```

Quick start

```r library(MLmorph)

set.seed(1) n <- 120 y <- factor(sample(letters[1:3], n, TRUE)) x1 <- factorizenumericvector(runif(n, 10, 20), breaksno = 3) x2 <- factorizenumericvector(runif(n, 1, 2), breaksno = 4) df <- data.frame(y, x1, x2)

1) Train a random forest and validate on a holdout

fit <- createrfmodel(df, dependent = "y", ntree = 100, trainvalidatesplit = 0.75) fit$model # randomForest object head(fit$variablesimportance) # importance per predictor fit$modelperformanceontest # caret::confusionMatrix ```

Morphospace: predicted probabilities over all configurations

```r ms <- create_morphospace(df, fit$model)

str(ms$morphospace)

Columns: predictors (x1, x2), '' (class label), 'calculated' (probability),

'purely_simulated' (TRUE if configuration not observed in df)

Example: top 10 configurations by predicted probability

head(ms$morphospace[order(ms$morphospace$calculated, decreasing = TRUE), ], 10)

Probability mass per class (sums to 1 for each unique predictor combination)

aggregate(calculated ~ x1 + x2, data = ms$morphospace, sum) ```

Factorization helpers

```r

Numeric → ordered factor via equal-frequency bins (default)

z <- runif(12) fz <- factorizenumericvector(z, breaks_no = 3)

Logical → factor with labeled levels

fb <- factorizebinaryvector(c(TRUE, FALSE, TRUE))

Character → factor with stable, numbered labels

fc <- factorizecharactervector(c("alpha", "beta", "alpha"))

Apply heuristics column-wise

df2 <- factorizenicelydataframe(data.frame( a = runif(20), b = c(TRUE, FALSE), c = c("x","y","x","z") )) str(df2) ```

Loading data from files

```r

CSV

tmpcsv <- tempfile(fileext = ".csv") utils::write.csv(data.frame(a = 1:3, b = c("x","y","z")), tmpcsv, row.names = FALSE) datcsv <- loaddata(tmp_csv)

JSON

tmpjson <- tempfile(fileext = ".json") jsonlite::writejson(list(a = 1:2, b = c("u","v")), tmpjson, autounbox = TRUE) datjson <- loaddata(tmp_json)

XLSX

tmpxlsx <- tempfile(fileext = ".xlsx") openxlsx::write.xlsx(data.frame(a = 1:2, b = c("m","n")), tmpxlsx) datxlsx <- loaddata(tmp_xlsx) ```

Shiny app

```r

Launch the interactive app from the installed package

(opens in browser; not run during automated checks)

MLmorph()

```

Reproducibility and citation

To ensure reproducibility, record: - the variable set and factorization (binning method and parameters), - model hyperparameters (e.g., ntree), - the random seed and train/validate split.

If you use MLmorph in research or consulting reports, please cite the package and your model specification.

License

MIT © Oskar Kosch. See LICENSE.md for details.

Issues

Please report bugs or feature requests at this repository's Issues page.