Im am trying to do a multi-step forecasting with multivariate time series, I have 9 variables (Y,X1,..X8) with 2270 samples for each variable, and I am trying to predict the future values of Y (70 future values).

I am wondering how far can i get a good accuracy? I used an lstm model but a get a very low accuracy: 15%!

I normalized my data and I am using a window size of 120.

I used the code provided by tensorflow documentation : https://www.tensorflow.org/tutorials/structured_data/time_series

my data: enter image description here

my predictions: enter image description here


    multi_step_model = tf.keras.models.Sequential()
multi_step_model.add(tf.keras.layers.LSTM(16, activation='relu'))

multi_step_model.compile(optimizer=tf.keras.optimizers.RMSprop(clipvalue=1.0), loss='mae')
  • 1
    $\begingroup$ Hi, unfortunately we don't have enough information about your task in order to help you. Please update your question in the following way: Add information on the dataset and the task; Add the code to reproduce the model and list all the models you tried so far; Explain how you chose to preprocess variables before feeding them into the model; Show some data about your model's performance. Whit that information we could help you. Thanks $\endgroup$ – Leevo Apr 27 at 7:42
  • $\begingroup$ Hello, I added the informations you requested... I hope you can help me $\endgroup$ – BalticOY Apr 27 at 9:07
  • $\begingroup$ Consider using another model. Neural nets don't do a good you job when the amount of data is small. In your case 2270 samples, seems very little. IMHO $\endgroup$ – lsmor Apr 27 at 9:25
  • $\begingroup$ Can you please, propose a model that can do a good job in my case? $\endgroup$ – BalticOY Apr 27 at 9:39
  • $\begingroup$ Btw, I saw many tutorials using 1000 samples for training, and they got good results! ? What do you think? $\endgroup$ – BalticOY Apr 27 at 9:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.