diff --git a/src/bindings/python/tests_compatibility/test_ngraph/test_reduction.py b/src/bindings/python/tests_compatibility/test_ngraph/test_reduction.py
index c5dd2fe54c0..e1bac52b738 100644
--- a/src/bindings/python/tests_compatibility/test_ngraph/test_reduction.py
+++ b/src/bindings/python/tests_compatibility/test_ngraph/test_reduction.py
@@ -142,7 +142,7 @@ def test_roi_align():
     [([5, 2], 0, False), ([5, 2], 1, False), ([5, 2, 6], 2, False), ([5, 2], 0, True)],
 )
 def test_cum_sum(input_shape, cumsum_axis, reverse):
-    input_data = np.arange(np.prod(input_shape)).reshape(input_shape)
+    input_data = np.arange(np.prod(input_shape), dtype=np.int64).reshape(input_shape)
 
     node = ng.cum_sum(input_data, cumsum_axis, reverse=reverse)
     assert node.get_output_size() == 1
diff --git a/src/bindings/python/tests_compatibility/test_ngraph/test_roll.py b/src/bindings/python/tests_compatibility/test_ngraph/test_roll.py
index a2446284e03..b912eb5ebc3 100644
--- a/src/bindings/python/tests_compatibility/test_ngraph/test_roll.py
+++ b/src/bindings/python/tests_compatibility/test_ngraph/test_roll.py
@@ -8,7 +8,7 @@ import numpy as np
 
 
 def test_roll():
-    input = np.reshape(np.arange(10), (2, 5))
+    input = np.reshape(np.arange(10, dtype=np.int64), (2, 5))
     input_tensor = ng.constant(input)
     input_shift = ng.constant(np.array([-10, 7], dtype=np.int32))
     input_axes = ng.constant(np.array([-1, 0], dtype=np.int32))
diff --git a/src/bindings/python/tests_compatibility/test_ngraph/test_sequence_processing.py b/src/bindings/python/tests_compatibility/test_ngraph/test_sequence_processing.py
index 9b5ba178753..e5e85480060 100644
--- a/src/bindings/python/tests_compatibility/test_ngraph/test_sequence_processing.py
+++ b/src/bindings/python/tests_compatibility/test_ngraph/test_sequence_processing.py
@@ -9,7 +9,14 @@ from ngraph.impl import Type
 
 def test_onehot():
     param = ng.parameter([3], dtype=np.int32)
-    model = ng.one_hot(param, 3, 1, 0, 0)
+    # output type is derived from 'on_value' and 'off_value' element types
+    # Need to set explicitly 'on_value' and 'off_value' types.
+    # If we don't do it explicitly, depending on OS/packages versions types can be unpredictably either int32 or int64
+    on_value = np.array(1, dtype=np.int64)
+    off_value = np.array(0, dtype=np.int64)
+    depth = 3
+    axis = 0
+    model = ng.one_hot(param, depth, on_value, off_value, axis)
     assert model.get_output_size() == 1
     assert model.get_type_name() == "OneHot"
     assert list(model.get_output_shape(0)) == [3, 3]
diff --git a/tests/layer_tests/mo_python_api_tests/test_mo_convert_tf.py b/tests/layer_tests/mo_python_api_tests/test_mo_convert_tf.py
index 08589e3cbc5..f83b75a78e7 100644
--- a/tests/layer_tests/mo_python_api_tests/test_mo_convert_tf.py
+++ b/tests/layer_tests/mo_python_api_tests/test_mo_convert_tf.py
@@ -375,7 +375,7 @@ def create_tf_stateful_partioned_call_net(temp_dir):
 
     param1 = ov.opset8.parameter(data_shape, dtype=np.float32)
     param2 = ov.opset8.parameter(filters_shape, dtype=np.float32)
-    transpose2 = ov.opset8.transpose(param2, np.array([3, 2, 0, 1]))
+    transpose2 = ov.opset8.transpose(param2, np.array([3, 2, 0, 1], dtype=np.int64))
     conv = ov.opset11.convolution(param1, transpose2, strides, pads_begin, pads_end, dilations, auto_pad="same_upper")
 
     parameter_list = [param1, param2]