Built a machine learning model to predict 12 ahead

2019년 8월 16일

Model predicts that bitcoin will be at a lower value than currently 12 hours ahead.

2019년 8월 17일

노트

The model is a sequential RNN (Recurrent neural network) made using TensorFlow. The model predicts processes data of the past 240 hours to predict if in 12 hours, the price will be higher or lower. Data: BTC_Close, BTC_Volume, ETH_Close, ETH_Volume, BTC_MA7, BTC_MA30 and the fear and greed index. alternative.me/crypto/fear-and-greed-index/
Training data starts from Feb 2018 until now.
On validation data, the models trained seem to reach 55%-60% accuracy.
This particular prediction was 98% confident that the price would decrease (WITH NO PARTICULAR TARGET PRICE)
There is no underlying traditional strategy influencing decisions, simply machine learning techniques applied to the data, to classify as buy or sell.