diff --git a/Project.toml b/Project.toml
index f41791a..2f1220c 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,7 +1,7 @@
 name = "UnsupervisedClustering"
 uuid = "c6b3b713-2a2b-4e07-a35c-d57ac4f46186"
 authors = ["raphasampaio", "joaquimg", "mvpoggi", "vidalt"]
-version = "0.4.6"
+version = "0.4.7"
 
 [deps]
 Distances = "b4f34e82-e78d-54a5-968a-f98e89d6e8f7"
diff --git a/docs/docs.bat b/docs/docs.bat
index 48934ec..debc154 100644
--- a/docs/docs.bat
+++ b/docs/docs.bat
@@ -1,6 +1,6 @@
 @echo off
 
-SET BASEPATH=%~dp0
+SET BASE_PATH=%~dp0
 
-CALL julia +1.12 --project=%BASEPATH% -e "using Pkg; Pkg.develop(PackageSpec(path=dirname(pwd()))); Pkg.instantiate()"
-CALL julia +1.12 --project=%BASEPATH% %BASEPATH%\make.jl
\ No newline at end of file
+CALL julia +1.12 --project=%BASE_PATH% -e "using Pkg; Pkg.develop(PackageSpec(path=dirname(pwd()))); Pkg.instantiate()"
+CALL julia +1.12 --project=%BASE_PATH% %BASE_PATH%\make.jl
\ No newline at end of file
diff --git a/docs/docs.sh b/docs/docs.sh
index 7942634..44554d6 100755
--- a/docs/docs.sh
+++ b/docs/docs.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 
-BASEPATH="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+BASE_PATH="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
 
 julia +1.12 --project -e "using Pkg; Pkg.develop(PackageSpec(path=dirname(pwd()))); Pkg.instantiate()"
-julia +1.12 --project $BASEPATH/make.jl
\ No newline at end of file
+julia +1.12 --project $BASE_PATH/make.jl
\ No newline at end of file
diff --git a/format/format.bat b/format/format.bat
index 77915b7..d16faf7 100644
--- a/format/format.bat
+++ b/format/format.bat
@@ -1,5 +1,5 @@
 @echo off
 
-SET BASEPATH=%~dp0
+SET BASE_PATH=%~dp0
 
-CALL julia +1.12 --project=%BASEPATH% %BASEPATH%\format.jl
\ No newline at end of file
+CALL julia +1.12 --project=%BASE_PATH% %BASE_PATH%\format.jl
\ No newline at end of file
diff --git a/format/format.sh b/format/format.sh
index 646c842..2621af5 100755
--- a/format/format.sh
+++ b/format/format.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
 
-BASEPATH="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+BASE_PATH="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
 
-julia +1.12 --project=$BASEPATH $BASEPATH/format.jl
\ No newline at end of file
+julia +1.12 --project=$BASE_PATH $BASE_PATH/format.jl
\ No newline at end of file
diff --git a/revise/revise.bat b/revise/revise.bat
index d12e24d..ffde339 100644
--- a/revise/revise.bat
+++ b/revise/revise.bat
@@ -1,5 +1,5 @@
 @echo off
 
-SET BASEPATH=%~dp0
+SET BASE_PATH=%~dp0
 
-CALL julia +1.12 --project=%BASEPATH% --interactive --load=%BASEPATH%\revise.jl
+CALL julia +1.12 --project=%BASE_PATH% --interactive --load=%BASE_PATH%\revise.jl
diff --git a/revise/revise.sh b/revise/revise.sh
index 3e815f0..96cd886 100755
--- a/revise/revise.sh
+++ b/revise/revise.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
 
-BASEPATH="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+BASE_PATH="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
 
-julia +1.12 --project=$BASEPATH --interactive $BASEPATH/revise.jl
\ No newline at end of file
+julia +1.12 --project=$BASE_PATH --interactive $BASE_PATH/revise.jl
\ No newline at end of file
diff --git a/src/metaheuristic/random_swap.jl b/src/metaheuristic/random_swap.jl
index 3103fe5..08db7f6 100644
--- a/src/metaheuristic/random_swap.jl
+++ b/src/metaheuristic/random_swap.jl
@@ -4,6 +4,7 @@
         verbose::Bool = DEFAULT_VERBOSE
         max_iterations::Int = 200
         max_iterations_without_improvement::Int = 150
+        discard_empty_clusters::Bool = false
     ) where {LS <: AbstractAlgorithm}
 
 RandomSwap is a meta-heuristic approach used for clustering problems. It follows an iterative process that combines local optimization with perturbation to explore the search space effectively. A local optimization algorithm is applied at each iteration to converge toward a local optimum. Then, a perturbation operator generates a new starting point and continues the search.
@@ -16,6 +17,7 @@ RandomSwap is a meta-heuristic approach used for clustering problems. It follows
 - `verbose`: controls whether the algorithm should display additional information during execution.
 - `max_iterations`: represents the maximum number of iterations the algorithm will perform before stopping.
 - `max_iterations_without_improvement`: represents the maximum number of iterations allowed without improving the best solution.
+- `discard_empty_clusters`: when `true`, candidate solutions whose local-search result contains at least one empty cluster are rejected and never replace the current best solution.
 
 # References
 * Fränti, Pasi.
@@ -27,6 +29,7 @@ Base.@kwdef struct RandomSwap{LS <: AbstractAlgorithm} <: AbstractAlgorithm
     verbose::Bool = DEFAULT_VERBOSE
     max_iterations::Int = 200
     max_iterations_without_improvement::Int = 150
+    discard_empty_clusters::Bool = false
 end
 
 @doc """
@@ -83,7 +86,9 @@ function fit(meta::RandomSwap{LS}, data::AbstractMatrix{<:Real}, k::Integer) whe
             print_result(result)
         end
 
-        if isbetter(result, best_result)
+        discard = meta.discard_empty_clusters && has_empty_clusters(result)
+
+        if !discard && isbetter(result, best_result)
             best_result = result
             iterations_without_improvement = 0
 
@@ -92,6 +97,11 @@ function fit(meta::RandomSwap{LS}, data::AbstractMatrix{<:Real}, k::Integer) whe
             end
         else
             iterations_without_improvement += 1
+
+            if meta.verbose && discard
+                print_string("(discarded: empty cluster)")
+            end
+
             if iterations_without_improvement > meta.max_iterations_without_improvement
                 break
             end
diff --git a/src/result/counts.jl b/src/result/counts.jl
index ac3352a..cdc14e1 100644
--- a/src/result/counts.jl
+++ b/src/result/counts.jl
@@ -1,3 +1,7 @@
 function counts(result::AbstractResult)
     return StatsBase.counts(result.assignments, result.k)
 end
+
+function has_empty_clusters(result::AbstractResult)
+    return any(==(0), counts(result))
+end
diff --git a/test/test.bat b/test/test.bat
index a290856..4d8681b 100644
--- a/test/test.bat
+++ b/test/test.bat
@@ -1,9 +1,9 @@
 @echo off
 
-SET BASEPATH=%~dp0
+SET BASE_PATH=%~dp0
 
 IF "%~1"=="" (
-    CALL julia +1.12 --project=%BASEPATH%\.. -e "import Pkg; Pkg.test()"
+    CALL julia +1.12 --project=%BASE_PATH%\.. -e "import Pkg; Pkg.test()"
 ) ELSE (
-    CALL julia +1.12 --project=%BASEPATH%\.. %BASEPATH%\runtests.jl %1
+    CALL julia +1.12 --project=%BASE_PATH%\.. %BASE_PATH%\runtests.jl %1
 )
\ No newline at end of file
diff --git a/test/test.sh b/test/test.sh
index 7c2e11d..4859d81 100755
--- a/test/test.sh
+++ b/test/test.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
 
-BASEPATH="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+BASE_PATH="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
 
-julia +1.12 --project=$BASEPATH/.. -e "import Pkg; Pkg.test()"
\ No newline at end of file
+julia +1.12 --project=$BASE_PATH/.. -e "import Pkg; Pkg.test()"
\ No newline at end of file
diff --git a/test/test_discard_empty_clusters.jl b/test/test_discard_empty_clusters.jl
new file mode 100644
index 0000000..5915a2b
--- /dev/null
+++ b/test/test_discard_empty_clusters.jl
@@ -0,0 +1,60 @@
+module TestDiscardEmptyClusters
+
+using Random
+using Test
+using UnsupervisedClustering
+
+Base.@kwdef struct EmptyingLocalSearch <: UnsupervisedClustering.AbstractAlgorithm
+    rng::AbstractRNG = MersenneTwister(0)
+end
+
+function UnsupervisedClustering.fit(::EmptyingLocalSearch, data::AbstractMatrix{<:Real}, k::Integer)
+    n, d = size(data)
+    result = UnsupervisedClustering.KmeansResult(zeros(Int, n), zeros(d, k))
+    for i in 1:n
+        result.assignments[i] = mod(i - 1, k) + 1
+    end
+    result.objective = 100.0
+    return result
+end
+
+function UnsupervisedClustering.fit!(
+    ::EmptyingLocalSearch,
+    ::AbstractMatrix{<:Real},
+    result::UnsupervisedClustering.KmeansResult,
+)
+    fill!(result.assignments, 1)
+    result.objective = 1.0
+    return nothing
+end
+
+@testset "RandomSwap - discard_empty_clusters" begin
+    data = rand(MersenneTwister(1), 50, 2)
+    k = 4
+
+    no_discard = RandomSwap(
+        local_search = EmptyingLocalSearch(rng = MersenneTwister(2)),
+        max_iterations = 5,
+        max_iterations_without_improvement = 100,
+        discard_empty_clusters = false,
+        verbose = true,
+    )
+    result_no_discard = UnsupervisedClustering.fit(no_discard, data, k)
+    @test count(==(0), UnsupervisedClustering.counts(result_no_discard)) == k - 1
+    @test result_no_discard.objective == 1.0
+
+    discard = RandomSwap(
+        local_search = EmptyingLocalSearch(rng = MersenneTwister(2)),
+        max_iterations = 5,
+        max_iterations_without_improvement = 100,
+        discard_empty_clusters = true,
+        verbose = true,
+    )
+    result_discard = UnsupervisedClustering.fit(discard, data, k)
+    @test count(==(0), UnsupervisedClustering.counts(result_discard)) == 0
+    @test result_discard.objective == 100.0
+
+    return nothing
+end
+
+end