Cryptocurrency Price Prediction

Developed a Bidirectional LSTM model for predicting next day closing prices with a MAPE of 19% and built a dynamic portfolio optimization algorithm incorporating profit-taking and stop-loss strategies.

Table of Contents

• Overview
• Introduction
• Solution Approach
  • Data Preprocessing
  • Exploratory Data Analysis (EDA)
  • Predictive Modeling
  • Portfolio Optimization
• Conclusion

Overview

• Developed a Bidirectional Long Short-Term Memory (BiLSTM) model for predicting next-day closing prices of cryptocurrencies.
• Achieved a Mean Absolute Percentage Error (MAPE) of 19%.
• Built a dynamic portfolio optimization algorithm incorporating profit-taking and stop-loss strategies to maximize returns and minimize losses.

Introduction

Cryptocurrencies are highly volatile, with 75% of Bitcoin investors and 90% of stock investors losing money. Investors face challenges in predicting returns and optimizing portfolios due to the market's dynamic nature. The solution aims to provide AI-driven predictions and optimized trading strategies to help investors make informed decisions.

Solution Approach

Data Preprocessing

• Merging Trade Data with Asset Data: Combined trading data with asset metadata for comprehensive analysis.
• Timestamp Conversion: Converted timestamps to a human-readable date-time format.
• Aggregation to Daily Level: Aggregated minute-level trading data to daily-level for analysis.
• Feature Engineering: Created additional features such as Simple Moving Averages (SMA) and Relative Strength Index (RSI).

Exploratory Data Analysis (EDA)

• Data Overview: Analyzed a dataset containing 24 million rows and 10 columns.
• Volatility Analysis: Identified Bitcoin as the most volatile asset with high standard deviation in closing prices.
• Trend Identification: Visualized trends and patterns in the dataset.

Predictive Modeling

• Sequential Neural Network: Implemented a Sequential Neural Network with ReLU activation and Adam optimizer.
• Bidirectional LSTM (BiLSTM) Model: Developed a BiLSTM model with 64 units in each layer and a dropout rate of 0.2 to prevent overfitting. The model effectively captured dependencies in both forward and backward directions of the time series.
• Model Evaluation: Achieved a Mean Absolute Percentage Error (MAPE) of 19%.

Portfolio Optimization

• Buy/Sell Logic: Created a buy/sell logic to optimize the portfolio and maximize results.
• Dynamic Strategies: Implemented dynamic stop-loss and take-profit strategies to manage risks.
• Performance Tracking: Tracked metrics such as wallet balance, portfolio current value, and asset performance daily.

Conclusion

The project showcases the application of advanced AI techniques to predict cryptocurrency prices and optimize investment portfolios. By leveraging a BiLSTM model and dynamic trading strategies, the solution aims to enhance investors' decision-making processes and improve their chances of profitability in the volatile cryptocurrency market.