diff --git a/R/EnsembleFSResult.R b/R/EnsembleFSResult.R
index 284d860c..1544cf26 100644
--- a/R/EnsembleFSResult.R
+++ b/R/EnsembleFSResult.R
@@ -126,11 +126,17 @@ EnsembleFSResult = R6Class("EnsembleFSResult",
     #' The weights used are equal to the performance scores of each voter/model (or the inverse scores if the measure is minimized).
     #' The un-weighted methods use same weights for all voters (equal to 1).
     #'
+    #' Note that some methods output a feature ranking instead of a score per feature.
+    #' Therefore we also calculate **Borda's score**:
+    #' \eqn{s_{borda} = (p-i)/(p-1)}, where \eqn{p} is the total number of features, and \eqn{i} is the feature ranking.
+    #' So the best feature gets a borda score of \eqn{1} and the worst-ranked feature a borda score of \eqn{0}.
+    #' This score is method-agnostic, i.e. it can be used to compare the feature rankings across different methods.
+    #'
     #' The following methods are currently supported:
     #'
     #' - `"av"|"av_weighted"` (approval voting) selects the candidates that have the highest approval score, i.e. the features that appear the most often.
     #' This is the default feature ranking method.
-    #' - `"sav"|"sav_weighted"` (satisfaction approval voting) selects the candidates that have a higher satisfaction score,in proportion to the size of the voters approval sets.
+    #' - `"sav"|"sav_weighted"` (satisfaction approval voting) selects the candidates that have a higher satisfaction score, in proportion to the size of the voters approval sets.
     #' Voters who approve more candidates contribute a lesser score to the individual approved candidates.
     #'
     #' @param method (`character(1)`)\cr
@@ -138,6 +144,8 @@ EnsembleFSResult = R6Class("EnsembleFSResult",
     #'
     #' @return A [data.table::data.table] listing all the features, ordered by decreasing scores (depends on the `"method"`).
     #' An extra column `"norm_score"` is produced for methods for which the original scores (i.e. approval counts in the case of approval voting) can be normalized and interpreted as **selection probabilities**, see Meinshausen et al. (2010).
+    #' The `"borda_score"` column is always included to incorporate feature ranking methods that don't output per-feature scores but only rankings.
+    #'
     feature_ranking = function(method = "av") {
       assert_choice(method, choices = c("av", "av_weighted", "sav", "sav_weighted"))
 
diff --git a/R/voting_methods.R b/R/voting_methods.R
index 2f69a1c8..a39ff547 100644
--- a/R/voting_methods.R
+++ b/R/voting_methods.R
@@ -1,9 +1,10 @@
 # some collection of voting methods for feature ranking
 
-# Parameters:
 # @param voters list of feature vectors (best features, each a subset of "candidates")
 # @param candidates vector with ALL features
 # @param weights vector of weights, 1-1 correspondence with voters
+# @return data.table with 4 columns: features (mandatory), score, norm_score, borda_score (mandatory). Always features are ordered with decreasing `score` (or descreasing
+# `borda_score` if a method returns only a ranking).
 
 approval_voting = function(voters, candidates, weights) {
   # faster R version in case of equal weights
@@ -31,10 +32,11 @@ approval_voting = function(voters, candidates, weights) {
     setorderv(res, cols = "score", order = -1)
   }
 
-  res
+  res[, borda_score := (nrow(res) - .I) / (nrow(res) - 1)]
 }
 
 satisfaction_approval_voting = function(voters, candidates, weights) {
   res = as.data.table(SAV_rcpp(voters, candidates, weights))
   setorderv(res, cols = "score", order = -1)
+  res[, borda_score := (nrow(res) - .I) / (nrow(res) - 1)]
 }
diff --git a/man/AutoFSelector.Rd b/man/AutoFSelector.Rd
index a7d46be4..9c59c626 100644
--- a/man/AutoFSelector.Rd
+++ b/man/AutoFSelector.Rd
@@ -149,7 +149,6 @@ Hash (unique identifier) for this partial object, excluding some components whic
 \if{html}{\out{
 <details><summary>Inherited methods</summary>
 <ul>
-<li><span class="pkg-link" data-pkg="mlr3" data-topic="Learner" data-id="estimate_memory_usage"><a href='../../mlr3/html/Learner.html#method-Learner-estimate_memory_usage'><code>mlr3::Learner$estimate_memory_usage()</code></a></span></li>
 <li><span class="pkg-link" data-pkg="mlr3" data-topic="Learner" data-id="format"><a href='../../mlr3/html/Learner.html#method-Learner-format'><code>mlr3::Learner$format()</code></a></span></li>
 <li><span class="pkg-link" data-pkg="mlr3" data-topic="Learner" data-id="help"><a href='../../mlr3/html/Learner.html#method-Learner-help'><code>mlr3::Learner$help()</code></a></span></li>
 <li><span class="pkg-link" data-pkg="mlr3" data-topic="Learner" data-id="predict"><a href='../../mlr3/html/Learner.html#method-Learner-predict'><code>mlr3::Learner$predict()</code></a></span></li>
diff --git a/man/ensemble_fs_result.Rd b/man/ensemble_fs_result.Rd
index a5978994..ade481fc 100644
--- a/man/ensemble_fs_result.Rd
+++ b/man/ensemble_fs_result.Rd
@@ -211,11 +211,17 @@ Most methods have a \code{"*_weighted"} version that outputs a weighted score.
 The weights used are equal to the performance scores of each voter/model (or the inverse scores if the measure is minimized).
 The un-weighted methods use same weights for all voters (equal to 1).
 
+Note that some methods output a feature ranking instead of a score per feature.
+Therefore we also calculate \strong{Borda's score}:
+\eqn{s_{borda} = (p-i)/(p-1)}, where \eqn{p} is the total number of features, and \eqn{i} is the feature ranking.
+So the best feature gets a borda score of \eqn{1} and the worst-ranked feature a borda score of \eqn{0}.
+This score is method-agnostic, i.e. it can be used to compare the feature rankings across different methods.
+
 The following methods are currently supported:
 \itemize{
 \item \code{"av"|"av_weighted"} (approval voting) selects the candidates that have the highest approval score, i.e. the features that appear the most often.
 This is the default feature ranking method.
-\item \code{"sav"|"sav_weighted"} (satisfaction approval voting) selects the candidates that have a higher satisfaction score,in proportion to the size of the voters approval sets.
+\item \code{"sav"|"sav_weighted"} (satisfaction approval voting) selects the candidates that have a higher satisfaction score, in proportion to the size of the voters approval sets.
 Voters who approve more candidates contribute a lesser score to the individual approved candidates.
 }
 }
@@ -223,6 +229,7 @@ Voters who approve more candidates contribute a lesser score to the individual a
 \subsection{Returns}{
 A \link[data.table:data.table]{data.table::data.table} listing all the features, ordered by decreasing scores (depends on the \code{"method"}).
 An extra column \code{"norm_score"} is produced for methods for which the original scores (i.e. approval counts in the case of approval voting) can be normalized and interpreted as \strong{selection probabilities}, see Meinshausen et al. (2010).
+The \code{"borda_score"} column is always included to incorporate feature ranking methods that don't output per-feature scores but only rankings.
 }
 }
 \if{html}{\out{<hr>}}