ANN classification - neural-network

Question

Using a small and imbalanced data set, I put the following codes before all other codes.
from numpy.random import seed
seed(10)
from tensorflow import random
random.set_seed(10)
After standardizing the features, I put
skf = StratifiedKFold(n_splits=10, random_state=10, shuffle=True)
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.4, random_state=10)
# define a base ANNC model
def create_model():
model = Sequential()
model.add(Dense(units=64, kernel_initializer='normal', activation='relu', input_shape=(n_features,)))
model.add(Dense(8, kernel_initializer='normal', activation='relu'))
model.add(Dense(4, kernel_initializer='normal', activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics='acc')
return model
After defining a ANN classification, and then I evaluate a base model.
base_model = KerasClassifier(build_fn=create_model, batch_size=10, epochs=1000, verbose=0)
Every time I run the base model, the results are the same. But when I use grid search, the grid_search.best_params_ and grid_search.best_score_ are different. I use the following codes:
# define a ANNC
def create_model():
model = Sequential()
model.add(Dense(units=64, kernel_initializer='normal', activation='relu', input_shape=(n_features,)))
model.add(Dense(8, kernel_initializer='normal', activation='relu'))
model.add(Dense(4, kernel_initializer='normal', activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics='acc')
return model
skf = StratifiedKFold(n_splits=10, random_state=10, shuffle=True)
# create the parameter grid
batch_size = [10,20,30,40,50,60,70,80,90,100]
epochs = [100,200,300,400,500,600,700,800,900,1000]
param_grid = dict(batch_size=batch_size, nb_epoch=epochs)
# create a multi-class ANNC
model = KerasClassifier(build_fn=create_model, verbose=0)
grid_search = GridSearchCV(estimator=model, param_grid=param_grid, n_jobs=-1, verbose=0)
# run the GridSearchCV process
grid_search = grid_search.fit(X_train, Y_train)
print("The best values of param_grid are:", grid_search.best_params_, grid_search.best_score_)
I found ANN is very difficult to do when compared to other algorithms. It would be much appreciated if any one could help me. Thanks in advance.