Multivariate LSTM Model to forecast data

Hi All, I am doing an assignment related to LSTM model, The model couldnt forecast the multivariate data(weekwise) for more that a 3weeks, It gets saturated after 3 weeks of forecast, Could you please help me what alternate things should I try?

I think your this has to do with your window size. Chan you check that your windows size is moving as you predict new forecasts!

1 Like

I have checked the window size is moving,I have created a csv file where the data forecasted is saved. and the same forecast data is used to predict the future value. I have attached the image of output.

I mean after each row is predicted and saved than new window is taken

I guess the model is getting saturated after 3 weeks of forecast. The plot contains a feature on y axis and weeks on x axis.

It seems to me that at the end of the data the model is predicting the data after the last window size, so you probably need to feed predicted data as actual data to the model to continue moving the window size, that’s what it seems to me but maybe the problem is elsewhere too.

Yaa. Your right. I have done the same. after last window i take the predicted data into new window and feed the data to predict next data. But the model is getting saturated after 3 to 4 predictions. It couldnt follow the trend