訓練データの正解率よりテストデータの正解率が高いのは何を指している？

Question

MNISTのデータを分割してドロップアウトを使い訓練データとテストデータを比較したのだが、分割せずに60000枚の時は妥当な結果が出た45,000枚に訓練データを減らした時訓練データの正解率よりテストデータの正解率のほうが高くなりました。これは一体どういう意味なのでしょうか？教えていただけると幸いです。
参考にコードを載せておきます。

import keras
from keras.datasets import mnist
import numpy as np
from sklearn.model_selection import train_test_split
import tensorflow as tf
from tensorflow.keras import datasets, layers, models
from sklearn.model_selection import StratifiedKFold
from keras.layers.core import Dropout
(x_train, y_train), (x_test, y_test) = mnist.load_data()
skf = StratifiedKFold(n_splits=4)# 1/4に分割
for train_index, test_index in skf.split(x_train,y_train):
 x_train_X = x_train[train_index]
 y_train_Y = y_train[train_index]
shape = (28, 28, 1)
x_train_X = x_train_X.reshape(-1,shape[0],shape[1],shape[2])
x_test = x_test.reshape(-1,shape[0],shape[1],shape[2])
print(x_train_X.shape)
print(y_train_Y.shape)
print(x_test.shape)
print(y_test.shape)
x_train_X, x_test = x_train_X / 255.0, x_test / 255.0
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28,1)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.summary()
model.add(layers.Flatten())
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10, activation='softmax'))
keras.layers.Dropout(0.5)
model.compile(optimizer='sgd',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
epochs=10
batch_size = 128
history = model.fit(x_train_X,y_train_Y,batch_size=batch_size, epochs=epochs)
test_loss, test_acc = model.evaluate(x_test,  y_test, verbose=2)