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.
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:
Exploratory Data Analysis (EDA):
Analyzed a dataset containing 24 million rows and 10 columns, focusing on trends and patterns.
Identified Bitcoin as the most volatile asset with a high standard deviation in closing prices.
Data Preprocessing:
Created a daily-level dataset from minute-level data.
Handled data anomalies and performed feature engineering, including scaling features and creating additional variables like RSI indicators and lagged features for close price and volume.
Predictive Modeling:
Implemented a Sequential Neural Network with ReLU activation and Adam optimizer.
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.
Portfolio Optimization:
Created a buy/sell logic to optimize the portfolio and maximize results.
Implemented dynamic stop-loss and take-profit strategies to manage risks.
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.