Enhancing Cryptocurrency Price Forecasting: a Comparative Analysis of Feature Quantity and Forecasting Models

EasyChair Preprint 11385

Abstract

Cryptocurrencies such as Bitcoin have become a popular digital currency in recent years, attracting investors and traders worldwide. However, the volatile nature of Bitcoin prices poses a challenge for predicting future prices accurately. To this end, this articles examines how different forecasting models for Bitcoin prices are affected by the number of features and which models are more effective at forecasting. We apply different ARIMA, SVM, LSTM, models on the Bitcoin historical dataset from Kaggle to predict the prices. With one-minute intervals, the dataset spans from 2012-01-01 to 2021-03-31. The data is separated into subsets for training (70%) and testing (30%) to get performance results of the models. The results show that the LSTM model performs better only when the appropriate features are selected. The study emphasizes the importance of selecting appropriate features to improve forecasting accuracy, as it can significantly impact the prediction of Bitcoin prices.

Keyphrases: 1. Cryptocurrency price forecasting, 2. Feature Selection, 3. Machine Learning

@booklet{EasyChair:11385,
  author    = {Mustabshera Fatima and Mir Sajjad Hussain Talpur and Zeeshan Ahmed Nizamani and Toufique Ahmed Nizamani and Muhammad Yaqoob Koondhar and Zulfikar Ahmed Maher},
  title     = {Enhancing Cryptocurrency Price Forecasting: a Comparative Analysis of Feature Quantity and Forecasting Models},
  howpublished = {EasyChair Preprint 11385},
  year      = {EasyChair, 2023}}