data

• KOSPI close price prediction for next day
• trained data KOSPI between '2014-01-01' to '2022-01-01'
• test data KOSPI between '2022-01-01' to '2022-07-01'

regression LSTM

• model structure (LSTM, {'n_layers':1, 'bidirectional': True,"rnn_dropout":0.3,"fc_dropout":0.9})
• mse loss 0.000741
• When use it as category classification, up down over 0.2% increase than prev close price
  • expected earning when only buy true predicted : 0.15% per day (without transaction fee(tax))
  • expected earning when all buy : -0.10% per day (without transaction fee(tax))

                precision    recall  f1-score   support   
  
         False       0.59      0.88      0.71     50376   
          True       0.46      0.14      0.21     35269   
  
      accuracy                           0.58     85645    
     macro avg       0.52      0.51      0.46     85645   
  weighted avg       0.54      0.58      0.51     85645   
  

• Java 11
• Spring boot 2.6.6
• JPA, hibernate
• Spring security
• mockito
• flyway
• lombok
• h2
• mariadb
• redis

• pinia
• billboard.js
• vue-sweetalert2

• tsai