Ethereum Price Prediction Comparison Using k-NN and Multiple Polynomial Regression

Abstract

Machine learning (ML) algorithms have been widely used to predict future financial trends. It has become a tool for predicting future trends based on what is known beforehand. Like other financial stock markets, cryptocurrency has become a new sensation and challenge for investors to predict its behaviour. However, unlike other financial instruments, cryptocurrency has been renowned because of the difficulty to predict the price due to its volatility behaviour that changes so rapidly and since there is no fundamental economy for its value. This paper presents a performance comparison of two ML algorithms in predicting Ethereum price with non-time series analysis, which are k- Nearest Neighbors (k-NN) and multiple polynomial regression (MPR). The experiment used independent variables from related real-world economic fundamentals such as Dow Jones Index, gold price, oil price, and Ethereum volume. The experiment data was collected from the records from April 2017 until April 2021. For each algorithm, several methods of preprocessing data were used to match all independent data with the dependent data. Three different preprocessing scenarios were also used to find the maximum accuracy model. scenario 1 (feature selection based on correlation matrix), scenario 2 (feature selection based on correlation with the dependent variables and among independent variables), and scenario 3 (scenario 1 extracted with PCA). The performance of the compared methods was evaluated by using MSE and MAE. From the experiment, a comparison of results using two different models with k-NN and multiple polynomial regression is obtained. It is found that k-NN with a hyperparameter K = 2 have the best prediction with MSE = 449.032 and MAE = 14.282 compared with multiple polynomial regression with the best MSE = 13953.96 and MAE = 84.923.