diff --git a/modules/ml/perf/perf_knn.cpp b/modules/ml/perf/perf_knn.cpp
new file mode 100644
index 0000000000..f911b6520e
--- /dev/null
+++ b/modules/ml/perf/perf_knn.cpp
@@ -0,0 +1,38 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+#include "perf_precomp.hpp"
+
+namespace opencv_test { namespace {
+
+// KNN brute-force findNearest: dominated by the per-sample L2 distance reduction.
+typedef TestBaseWithParam< tuple<int, int, int> > KNNFindNearest;  // (train samples, dims, K)
+
+PERF_TEST_P(KNNFindNearest, brute_force, testing::Values(
+            make_tuple(5000, 128, 5),
+            make_tuple(10000, 64, 10)))
+{
+    const int nsamples = get<0>(GetParam());
+    const int dims     = get<1>(GetParam());
+    const int K        = get<2>(GetParam());
+    const int nquery   = 2000;
+
+    Mat train(nsamples, dims, CV_32F), responses(nsamples, 1, CV_32F), query(nquery, dims, CV_32F);
+    RNG& rng = theRNG();
+    rng.fill(train, RNG::UNIFORM, 0.f, 1.f);
+    rng.fill(query, RNG::UNIFORM, 0.f, 1.f);
+    for (int i = 0; i < nsamples; i++)
+        responses.at<float>(i) = (float)(i & 1);
+
+    Ptr<KNearest> knn = KNearest::create();
+    knn->setAlgorithmType(KNearest::BRUTE_FORCE);
+    knn->setDefaultK(K);
+    knn->train(train, ROW_SAMPLE, responses);
+
+    Mat results, neighbors, dists;
+    TEST_CYCLE() knn->findNearest(query, K, results, neighbors, dists);
+
+    SANITY_CHECK_NOTHING();
+}
+
+}} // namespace
diff --git a/modules/ml/perf/perf_main.cpp b/modules/ml/perf/perf_main.cpp
new file mode 100644
index 0000000000..ccdf300c6e
--- /dev/null
+++ b/modules/ml/perf/perf_main.cpp
@@ -0,0 +1,6 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+#include "perf_precomp.hpp"
+
+CV_PERF_TEST_MAIN(ml)
diff --git a/modules/ml/perf/perf_precomp.hpp b/modules/ml/perf/perf_precomp.hpp
new file mode 100644
index 0000000000..de2c7acaba
--- /dev/null
+++ b/modules/ml/perf/perf_precomp.hpp
@@ -0,0 +1,15 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+#ifndef __OPENCV_PERF_PRECOMP_HPP__
+#define __OPENCV_PERF_PRECOMP_HPP__
+
+#include <opencv2/ts.hpp>
+#include <opencv2/ml.hpp>
+
+namespace opencv_test {
+using namespace perf;
+using namespace cv::ml;
+}
+
+#endif
diff --git a/modules/ml/src/knearest.cpp b/modules/ml/src/knearest.cpp
index bd9137b24b..c4aba028bc 100644
--- a/modules/ml/src/knearest.cpp
+++ b/modules/ml/src/knearest.cpp
@@ -41,6 +41,7 @@
 //M*/
 
 #include "precomp.hpp"
+#include "opencv2/core/hal/intrin.hpp"
 #include "kdtree.hpp"
 
 /****************************************************************************************\
@@ -171,13 +172,21 @@ public:
                 const float* u = _samples.ptr<float>(testidx + range.start);
 
                 float s = 0;
-                for( i = 0; i <= d - 4; i += 4 )
+                i = 0;
+#if (CV_SIMD || CV_SIMD_SCALABLE)
                 {
-                    float t0 = u[i] - v[i], t1 = u[i+1] - v[i+1];
-                    float t2 = u[i+2] - v[i+2], t3 = u[i+3] - v[i+3];
-                    s += t0*t0 + t1*t1 + t2*t2 + t3*t3;
+                    const int vl = VTraits<v_float32>::vlanes();
+                    v_float32 a0 = vx_setzero_f32(), a1 = vx_setzero_f32();
+                    for( ; i <= d - 2*vl; i += 2*vl )
+                    {
+                        v_float32 d0 = v_sub(vx_load(u + i), vx_load(v + i));
+                        a0 = v_fma(d0, d0, a0);
+                        v_float32 d1 = v_sub(vx_load(u + i + vl), vx_load(v + i + vl));
+                        a1 = v_fma(d1, d1, a1);
+                    }
+                    s = v_reduce_sum(v_add(a0, a1));
                 }
-
+#endif
                 for( ; i < d; i++ )
                 {
                     float t0 = u[i] - v[i];