In this experiment we predict the price movements of the next 8 hours. A positive value means, that the average price of the next 8h is expected to be higher than the average of the last 8h.
The predictions do not guarantee profits. They should only serve as an illustrative decision-making aid. We are not liable for damages or loss of money. Trading involves risks. Do not invest money that you need.
When predicting we are updating every 5 minutes. Bookmark us.
In this atypical construction we use a convolutional neural network. As input, the NN receives images with encoded RSI and MFI values.
The RSI/MFI parameters range from 45 minutes to 24 hours. Each image has a width of 480 pixels, which is translated to 8 hours. The corresponding class indicates whether the average price is higher or lower in the following 8 hours. It is compared with the average price of the sample.
The learning data consists of Bitcoin pairs between 2017 and 2018 and contains no overlaps. To improve the quality of the training data, samples with gaps were sorted out. A total of 104.753 training images and 81.460 validation images were created.
The model was trained with the Tensorflow gpu backend. During training, the validation protocol achieved an accuracy of 69% to 71%. That said, it is up to you to judge how well the application works in the real world.
The power indicator uses an algorithm that gives more weight to newer values. The power is used to sort the coin list. This is to show the latest advances and promising coins at the top.
Could you use this data for something interesting? Have you already made any experiments on this topic or would you like to exchange information with us? Feel free to contact us.
This is only a side project. You may be wondering what our core business is.
Head to our main page and find out more.