RNN not learning Keras

I have been training RNNs in keras for some time now, but recently I was faced with a problem of RNN not learning anything. Therefore, I want to make sure that I am feeding in the data correctly. I know that keras expect your input shape to be a 3-D tensor, so following is the code that I am using to reshape a 2-D numpy array to a 3D array. Please check if this is the right approach or not? The idea here is to take the first 3 input values and add it to the time dimension. I have also included a small snippet of the training session. You can see that the loss is changing extremely slowly after each epoch and the accuracy not improving at all.

My question here is this, am I making any mistake in reshaping the input? Also what could be the reasons for my model not learning anything? I have tried a lot of hyperparameters, optimizers, different batch sizes, simplified model as much as possible (just used a single RNN cell). Still nothing, no learning at all. This a classification problem, by the way, I have 118 features and 17 classes. I am using categorical cross entropy loss and Adam optimizer.

X_train_seq = []
for i in range(3,len(X_train_scaled)):          
     X_train_seq.append(X_train_scaled[i-3:i,0:])
X_train_seq = np.array(X_train_seq)
3 is the time step  

screenshot of the model getting stuck and not learning anything during training