CryptoHFT - Trading Algorithm
George Triafylos
Using machine learning to predict the stock market
The CryptoHFT aims to learn patterns in price changes that can help make predictions on future price movements of different cryptocurrencies. Combining convolutional neural networks, long short-term memory cells (LSTM’s), and fully connected layers, the CryptoHFT constantly trains over new datapoints to optimise how it interprets incoming price sequences.
Page Views:
The CryptoHFT is a machine learning model that aims to make future predictions of cryptocurrency price movements. Through the use of historical prices, the model learns the behaviour and patterns of the movements so that it can make reasonable predictions on where the next price point will be. The model is considered to be profitable as long as the prediction of the direction of the price movement has an accuracy over 50%.
The approach to design a suitable model began with research on what sort of network architectures, layers, and parameters are typically used in similar scenarios. These configurations were then set up to be used in a grid search algorithm that trains a model with all variations that required consideration.
The final optimal model correctly predicts the future price points with an average accuracy of 54.16%. Though this number is only slightly above 50%, statistically, it will still make small incremental profits that in the long term can eventuate into significant earnings.