How to improve the accuracy of LSTM network?

Hello everyone, I hope you're fine and safe

I am working on forecasting time-series data using LSTM network, but the forecasting of the test data seems very inaccurate.

Kindly find the attached code with the dataset.

Any suggestions to improve the accuracy please?

Another question is how can I modify this code to forecast a variable based on multible variables (i.e., Multivariate Multi-Step LSTM Models; Multiple Input Multi-Step Output)?

Thanks in advance,

[MatLabMcLovinPotato]

Hey Mohamed! The chicken pox timeseries example might help as it does what you're asking about. Could experiement with the options, layer config, and traiing from this, then adjust to suit your needs once you're able to get past the prediction flatlining (had the same issue and going back to this example really helped)

https://au.mathworks.com/help/deeplearning/ug/time-series-forecasting-using-deep-learning.html

[Sinan Islam]

Unfortunately, Matlab dont have a single example of optimizing LSTM.

[Bhavick Singh]

Hi Mohammed, Im having the same problem as well. Were you able to find the solution?

Answer 1

[Madderla Chiranjeevi]

Hi sirs, I am also strucking here, please help me in this regard if you solved this problem.you can mail me also over [e-mail]

I am so thankful for you for this

Answer 2

[yanqi liu]

yes，sir，may be use bilstmLayer to replace lstmLayer，and not use predictAndUpdateState during test

set randperm to shuffle data before make numTimeStepsTrain to split data

[omar]

unfortunatly biLSTM improve only the speed of training