NEWS

---

modelimpact 1.2.0

New features

• Cost and value arguments now accept per-observation values, not just a single number. tp_val and var_cost in cost_revenue(), profit(), roi(), marginal_profit(), break_even() and impact_summary(), and the per-cell value arguments (tp_val, fp_val, tn_val, fn_val) in profit_thresholds(), may be given as an unquoted column name (or a vector) so that each case can carry its own value or cost. A single scalar continues to work exactly as before.
• New budget-constrained targeting helpers: budget_profit() reports the best profit achievable for a given budget, and budget_frontier() sweeps a range of budgets to show how achievable profit grows with spend (with an autoplot() method and plot_budget() wrapper).
• Optional bootstrap confidence bands on the ranking curves. profit(), roi(), cost_revenue(), cumulative_gains() and lift_curve() gained ci, n_boot and conf_level arguments; with ci = TRUE the result gains .lower/.upper columns and autoplot() draws a ribbon.
• New support for regression models that predict a continuous value: value_gains() (with a concentration / Gini coefficient) and value_profit() rank cases by predicted value and report the share of realised value captured and the profit of targeting the top share. value_gains() has its own autoplot() method and plot_value_gains() wrapper.
• New support for uplift models evaluated against treated / control data: qini_curve() reports the cumulative incremental positives with a Qini coefficient and AUUC, and uplift_profit() turns the ranking into incremental profit. Both have autoplot() methods (plot_qini()).
• tidy() and glance() methods (via the generics package) let result objects flow into tidyverse / tidymodels pipelines. tidy() returns the result as a plain tibble; glance() gives a one-row headline summary for profit(), profit_thresholds(), budget_frontier(), value_gains() and qini_curve().
• New impact_report() renders a ready-made, parameterised quarto report bundling the headline summary and every curve into a single HTML document.

Documentation

• The package is now framed around classification impact in general, with customer churn as one example. A new "Beyond churn" vignette works through a fraud-detection use case.

modelimpact 1.1.0 (2026-07-08)

New features

• New ranking-based views: cumulative_gains(), lift_curve(), marginal_profit(), break_even() and impact_summary().
• New threshold / classification views: confusion_payoff(), roc_pr() and payoff_grid().
• New comparison and robustness tools: compare_models(), bootstrap_profit() and tornado().
• All result objects now have ggplot2::autoplot() methods, plus matching plot_*() helper functions, for one-line visualisation.
• New run_app() launches an interactive shiny application to explore the package and upload your own data (comma- or semicolon-separated CSV).
• profit(), cost_revenue() and roi() gained a positive argument so the event class no longer has to be labelled "Yes", and a prob_accept argument that reconciles them with profit_thresholds().

Bug fixes

• profit_thresholds() now correctly applies the fp_val (false-positive value) argument, which was previously ignored.

modelimpact 1.0.0 (2021-05-06)

• Added functionality for finding optimal thresholds