Training accuracy : 95% and Test accuracy : 35%

..................................................sdadasda...............................................

This looks like a typical overfitting problem. Did you try to reduce your model or to train on more data?

I have been working as data scientist from last 4 years, had experienced this scenario on many instances during beginning phase of my career. It is just game play between bias variance. Let me know if I can be any help

plz share the details of the data set. only then I will be able comment on the reasons of over fitting. use validation set to stop training if it decreas continuously. thanks and regards

I can prevent the over-fitting problem by changing LSTM to Bi-LSTM or GRU if the data is not too big. Also data augmentation is the best idea Relevant Skills and Experience I am a data scientist in NLP and deep learning. I used LSTM for machine translation, sequence classification and time series. Proposed Milestones $255 CAD - As I don't know the problem (sequence classification... etc), I will determine the milestones after What's the data type and number of samples?