diff --git a/benchmark/benchmark.r b/benchmark/benchmark.r
index dc66061..ad1bc8f 100644
--- a/benchmark/benchmark.r
+++ b/benchmark/benchmark.r
@@ -1,6 +1,8 @@
 library(glue)
 Sys.setenv(LANG = "en")
-resoutput <- capture.output(Sys.setlocale("LC_MESSAGES", 'en_GB.UTF-8'))
+if (Sys.getlocale("LC_MESSAGES") == "es_ES.UTF-8") {
+  resoutput <- capture.output(Sys.setlocale("LC_MESSAGES", 'en_GB.UTF-8'))
+}
 args = commandArgs(trailingOnly=TRUE)
 if (length(args)!=3) {
   stop("Only two arguments must be supplied (score & input_file & visualize).n", call.=FALSE)
diff --git a/benchmark/tests/Models_test.py b/benchmark/tests/Models_test.py
index ec66e48..ea5b12a 100644
--- a/benchmark/tests/Models_test.py
+++ b/benchmark/tests/Models_test.py
@@ -90,5 +90,16 @@ class ModelTest(TestBase):
             #     score_expected,
             #     score_computed,
             # )
-            self.assertSequenceEqual(Models.get_complexity(key, clf), value)
+            # Fix flaky test
+            if key == "AdaBoostStree":
+                # computed values
+                a_c, b_c, c_c = Models.get_complexity(key, clf)
+                # expected values
+                a_e, b_e, c_e = value
+                for c, e in zip((a_c, b_c, c_c), (a_e, b_e, c_e)):
+                    self.assertAlmostEqual(c, e, delta=0.25)
+            else:
+                self.assertSequenceEqual(
+                    Models.get_complexity(key, clf), value
+                )
             self.assertEqual(score_computed, score_expected, key)