Recurrent Neural Network gives the same output

Hi,

I tried to use Kears-RNN in Rapid Miner Studio.
I have a total amount of 9000 datasets where the datasets were split into ratio 80% to 20%. There are 4 input parameters and I need to predict an output.
In Keras, the procedures were Core layer (Reshape: to (1,4))--->Recurrent layer （LSTM, activation: tanh, recurrent activation: hard sigmoid)--->Core layer (Dense: Relu)---> Core layer (Dense; Linear)
I have tried using different activation method, but the output that I obtained still showing the same value.
Anyone can help on this issue? Thank you very much.