- Abinhav Sagar used a Long Short-Term Memory neural network to make the predictions.
- Sagar posted the four steps to these predictions in a blog post and GitHub.
The cryptocurrency industry has a reputation for being volatile, unpredictable, and ever-changing. Predicting the way that the market would move could easily give an advantage to the everyday investor, and one data scientist believes that he’s figured it out. Abinhav Sagar of the prestigious Vellore Institute of Technology recently stated that it is possible to use a Long Short-Term Memory (LSTM) neural network to predict these prices with real-world accuracy. Sagar published a blog about this exact method on December 2nd, showing the four steps he can take with the technology to create predictions in a “relatively unpredictable” market.
The demonstration started with a comment from Sagar that the application of this machine learning tech has been relatively limited in the cryptocurrency sector, even though it has had some success in the stock market. The cryptocurrency market prices like Bitcoin tend to move quickly, Sagar explains, as they correspond with changes in technology, security, politics, and economies around the world.
The four steps include:
- Collecting data about cryptocurrencies in real-time
- Preparing to use the data in neural network training
- Using the LSTM neural network to test the prediction
- Visualizing the prediction’s results
Sagar uses one of CryptoCompare’s datasets to train the network, applying the price, volume, open values, high values, and low values. The completed project can be found on GitHub, which the blog includes a link to. The page also details the functions that he applied to normalize the data values, in an effort to prepare the data for the neural network training.
Before reaching the fourth step to visualize the results, Sagar states that the evaluation metric he applied was Mean Absolute Error, which examines the average magnitude of the prediction sets’ errors. It measures this magnitude without any consideration for the direction.
Apart from market predictions, the use of decentralized tech (like blockchain) in conjunction with machine learning has been gaining momentum. NASA, for example, recently listed a data scientist position available, adding that expertise in the cryptocurrency and blockchain industries to be “a plus.” Furthermore, the agency is seeking out an individual that has qualifications in at least one of the following fields – big data, Internet of Things, analytics, cloud computing, machine learning, and statistics.