• Data model

  • Hidden layer -> why 2
  • Units - 32/16 -> next stage goal
  • Activation -> sigmoid -> classic + binary
  • Optimization -> gradient descent
  • Learning rate -> adaptive -> next stage
  • Flow diagram

• Result = 84% accuracy + 82% validation accuracy

  • plot accuracy vs iteration

Possible additional classification

• Predict the percentage of rise/fall

Possible additional features:

• NYSE/NASDAQ(or SP500)kind of reference on that day or days before
  • possibly, use the reference for a kind of stocks, like technology
• The performance prior to the day (2~3 days)
• Quadratic terms
• Other stock performance - in the same category

• Need to fine tune the stochastic training