我正在尝试构建入侵检测系统（IDS），并预测标签是良性还是DDoS攻击。但在各个Epoch中，我得到的准确率始终相同。

代码：

        from tensorflow import keras    import numpy as np    import datetime    import time    from tensorflow.keras.optimizers import Adam    from keras.models import Sequential    from keras.layers import Dense, Dropout    from keras import callbacks    x=pd.DataFrame(X)    x = x.values    sample = x.shape[0]    features = x.shape[1]    #Train: convert 2D to 3D for input RNN    x_train = np.reshape(x,(sample,features,1)) #shape  = (125973, 18, 1)    #Test: convert 2D to 3D for input RNN    x_test=pd.DataFrame(X_test)    x_test = x_test.values    x_test = np.reshape(x_test,(x_test.shape[0],x_test.shape[1],1))Model = keras.Sequential([        keras.layers.LSTM(80,input_shape=(features,x_train.shape[2]),                          activation='sigmoid',recurrent_activation='hard_sigmoid'),        keras.layers.Dense(1,activation="softmax")    ])Model.compile(optimizer='rmsprop',loss='mse', metrics=['accuracy'])#Training the modelModel.fit(x_train, y, epochs=10, batch_size= 32) Model.summary()# Final evaluation of the modelscores = Model.evaluate(x_test, y_test, verbose=0)print('/n')print("Accuracy: %.2f%%" % (scores[1]*100))Epoch 1/101074/1074 [==============================] - 92s 83ms/step - loss: 0.4180 - accuracy: 0.5820Epoch 2/101074/1074 [==============================] - 79s 74ms/step - loss: 0.4180 - accuracy: 0.5820Epoch 3/101074/1074 [==============================] - 81s 76ms/step - loss: 0.4180 - accuracy: 0.5820

解决方案是什么？

回答：

因为使用1个神经元的"softmax"激活函数总是输出1。你的神经元无法调整其输出以降低损失；从数学上讲，它只能返回1。