πŸ”¬
Minerva AI (EN)
  • πŸ‘‹Welcome to Minerva AI
  • STUDIES & FINDINGS
    • ☝️Abstract
    • πŸ“Introduction and related works
    • πŸ’‘Model overview
      • ♾️Convolutional neural network - CNN
      • ✍️Dropout techniques
      • 🧠Long short-term memory (LSTM)
      • πŸ™†Methodology
      • πŸ”—Data collection and merging
      • πŸ› οΈModel architecture
      • βž•Dataset and Parameter Optimization
      • πŸ‘ŒResults and discussion
      • βš“Trading Philosophy & Method
      • βž—Basic algorithm
      • 🌠Results
    • πŸ‘‰Conclusion
  • PRODUCT DEVELOPMENT
    • πŸ‘€Vision and development roadmap
    • 🌟Revenue models
    • πŸͺ™Tokenomics
  • TEAM
    • πŸ‘₯Founding team
  • RESOURCES
    • πŸ“—References
Powered by GitBook
On this page
  1. STUDIES & FINDINGS
  2. Model overview

Methodology

We propose a stacked CNN-LSTM model for predicting the price of a future market contract on the Binance cryptocurrency exchange, specifically the Solana (SOL/USDT) token. Although optimized for the cryptocurrency market, it can also be trained on other data. In image processing, where CNNs are widely used, 3x3 and 5x5 filters are commonly employed to capture spatial variance over a rectangular array of pixels. The input size of the kernel is modified to 1x54 to match the input feature set consisting of a total of 54 input features. Each input feature is represented by a series of variables such as closing price, technical indicator values, candlestick patterns, etc. These inputs are fed to the CNN to extract features related to daily price levels and generate a tensor that is then passed to the LSTM network. The LSTM takes inputs from the preceding 30 time steps (measured in days or hours) and attempts to predict the next time step of the data.

PreviousLong short-term memory (LSTM)NextData collection and merging

Last updated 1 year ago

πŸ’‘
πŸ™†