I have LSTM trading AI model for prediction of future price.
I have 10,000 data in training and separated into training (60%), validation (20%) and testing (20%). This 10,000 data is for years before 2022. The training went well and test result is 100%. This test dataset is not used in training.
Then test the model with new data downloaded from server for those data for 2023. But accuracy is very bad 30%. It was 100% for test dataset with <2022 dataset. How can I improve the model?