Scholarlink

This study addresses the challenges of predicting cryptocurrency prices by analyzing two prominent machine learning methodologies: support vector machine (SVM) models and artificial neural networks (ANN). Various SVM kernels and ANN architectures are explored to evaluate their forecasting performance. Using a dataset from October 1, 2020, to September 30, 2023, extensive experimentation and evaluation reveal the performance metrics (RMSE, MSE, MAE, and R2) for training and testing datasets of ethereum (ETH) and binance coin (BNB). Results show that SVM with a quadratic kernel (SVM2) consistently outperforms other SVM models. Among ANN models, those with narrow (ANN1) and medium-sized (ANN2) architectures demonstrate robust performance. Effective prediction methods were developed through iterative model refinement, parameter tuning, and k-fold cross-validation, emphasizing the importance of model simplicity and parameter optimization. These findings enhance the understanding of cryptocurrency market dynamics and enable more informed decision-making in this evolving domain.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning Insights into Cryptocurrency Price Prediction: SVM and ANN Perspectives

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sara Salehi

摘要

This study addresses the challenges of predicting cryptocurrency prices by analyzing two prominent machine learning methodologies: support vector machine (SVM) models and artificial neural networks (ANN). Various SVM kernels and ANN architectures are explored to evaluate their forecasting performance. Using a dataset from October 1, 2020, to September 30, 2023, extensive experimentation and evaluation reveal the performance metrics (RMSE, MSE, MAE, and R2) for training and testing datasets of ethereum (ETH) and binance coin (BNB). Results show that SVM with a quadratic kernel (SVM2) consistently outperforms other SVM models. Among ANN models, those with narrow (ANN1) and medium-sized (ANN2) architectures demonstrate robust performance. Effective prediction methods were developed through iterative model refinement, parameter tuning, and k-fold cross-validation, emphasizing the importance of model simplicity and parameter optimization. These findings enhance the understanding of cryptocurrency market dynamics and enable more informed decision-making in this evolving domain.