Bitcoin Cash price prediction using ML (SVM )

Santosh Roy
3 min readMar 2, 2021

Bitcoin cash (BCH) is an upgraded version of the original Bitcoin (BTC) blockchain, with variable block size between 8 to 32 MB. Increased block size enables fast and cheap BCH transactions in the network. However, this is not reflected in its market cap of about 9 billion USD. The Bitcoin on the other hand is steadily growing and recently broke 1 trillion USD market cap. Very much undervalued, the price of 1 BCH sits around 1/100th of a Bitcoin. The price ratio is largely driven by investors and institutional entities who has put more trust in Bitcoin. The best feature of BCH is that it can handle around 14k TPS, much higher than that of BTC. This is an asset which will be the primary deciding factor for the cryptocurrency adoption in the near future.

While the low price of BCH has no effect on the transaction speed, it is an important factor for investors and established financial houses. BCH price has shown stagnancy over the years, and even in the 2021 bull run, it has adhered to the BTC price trend. It has around 77% correlation with the BTC price and has little resilience to the tanking BTC price. Large fluctuations of BCH price has remained a cause of concern of its supporters and hodlers.

In this article I will cover some preliminary ML analysis I did using sklearn in python. One year data from 01/03/2020 to 28/02/2021 and using ML, I have some interesting insights to share!

Analysis details:

Using regression using radial basis function:

SVR parameters: (C=1000.0, cache_size=200, coef0=0.0, degree=3, epsilon=0.1, gamma=1e-05, kernel=’rbf’, max_iter=-1, shrinking=True, tol=0.001)

Data range: 01/03/2020 to 28/02/2021

Training data: 80 % (for set 1) and (90% for set 2)

Accuracy obtained: 0.72 (set 1) and 0.8 (set 2)

Results:

Plot 1: ML closing price prediction for BCH with 80% training data from 01/03/2020 to 28/02/2021
Plot 2: ML closing price prediction for BCH with 90% training data from 01/03/2020 to 28/02/2021

Discussion:

The two results with 80% and 90% of the training data set have some common features. Both of them shows a high around the 11/03/2021. In between there are regular highs and lows, the climb is also slow. This shows that the regular fluctuations are a part of the network. This is why DCA or hodl is good stategy in this type of volatile market (BCH volatility is around 6% daily). It is quite impossible in this scenario to profit by doing day trading.

15/03/2021 onwards it predicts a dip of ~15% to 40%. This an important test-point for this model to be valid. On the 25th the price should be somewhere around 575 to 650 USD range.

This model has used some basic info for prediction. There are number of other factors that can be used to build a better robust model. BTC Market cap, BTC-BCH correlation, and google trends for the BCH can be incorporated in the next iteration.

--

--

Santosh Roy

Dr Roy is an expert in nuclear physics & instrumentation. He is also a crypto enthusiast and analyst.