I implemented LSTM model seems like its predicting the trend correctly but underestimating all the time. Is there any way to fix it?
I am using multiple feature values in future and the target value in t-1 to predict. I am using sequence length =7 to forecast and the result is like below:
model used:
def LSTM_model():
activation = 'linear'
inputs = keras.Input(shape=input_shape)
x = layers.LSTM(512, activation=activation)(inputs)
x = layers.Dense(2 * 1024)(x)
output = layers.Dense(1)(x)
model = Model(inputs=inputs, outputs=output)
model.compile(optimizer='Adam', loss='mse', metrics=[r2_score])
return model
