我用角膜建立了一个经典的人工神经网络,它提供了结果(0或1)的概率(使用S型函数)。当模型适合〜90%时,模型的准确性很高,而测试集结果的结果概率非常低。我该怎么解释?
classifier = Sequential()
classifier.add(Dense(activation="relu",input_dim=7,kernel_initializer="uniform", units = 4))
classifier.add(Dense(activation="relu",kernel_initializer="uniform", units = 4))
classifier.add(Dense(activation="sigmoid", kernel_initializer="uniform", units = 1))
classifier.compile(optimizer="adam", loss="binary_crossentropy",metrics=['accuracy'])
classifier.fit(X_train,y_train, batch_size=10,epochs=100)
y_pred = classifier.predict(X_test)