Comparison of Support Vector Regression and Extreme Learning Machine Methods for Predicting Bitcoin Prices

Bitcoin can be used for transactions, mining, and investments. Transactions with Bitcoin are highly secure with the help of Bitcoin miner validation. Miners who validate transactions are rewarded with Bitcoins which then adds supply to the Bitcoin network. However, over time, these rewards will run out. The depletion of Bitcoin supply can affect the price of Bitcoin. In addition, investing in Bitcoin is very risky with the fluctuating price of Bitcoin. Therefore, it is necessary to predict the price. In this research, prediction is done using Support Vector Regression (SVR) and Extreme Learning Machine (ELM). The dataset for Bitcoin price (USD) comes from Yahoo Finance. The types of Bitcoin prices predicted are Open, High, Low, and Close prices. Across all series and both splits, ELM outperforms SVR. Under the 80/20 split, the average error of ELM is MAE 418.698 USD, RMSE 633.953 USD, R² of 0.987, versus SVR’s MAE 1061.449 USD, RMSE 1227.499 USD, R² of 0.955. A reduction of 60.6% (MAE) and 48.4% (RMSE). With the 60/40 split, ELM remains strong (MAE 550.783 USD, RMSE 850.656 USD, R² 0.989 while SVR deteriorates (MAE 1843.534 USD, RMSE 2093.542 USD, R² of 0.935, yielding 70.1% and 59.4% average reductions in MAE and RMSE, respectively. ELM consistently tracks both levels and day to day movements, with typical errors of only a few hundred dollars. These results indicate that ELM is the more reliable choice and is capable of capturing non-linearities for Bitcoin price prediction.