Scholarlink

Cryptocurrencies, particularly Ethereum (ETH), have significantly influenced the global economy, offering numerous investment opportunities. However, their inherent market volatility presents challenges for accurate forecasting. This study investigates the application of multiheaded artificial neural networks (ANNs) for predicting ETH prices. To improve forecasting performance, this work introduces a modified metaheuristic optimization algorithm, a variant of the particle swarm optimization (PSO) algorithm. A comparative analysis is conducted using a publicly available dataset to evaluate the model’s effectiveness. The modified algorithm consistently delivers superior results, optimizing the multiheaded ANN model to achieve a mean squared error (MSE) of 0.43908. Collected outcomes indicate that the proposed model is a viable solution for real-world cryptocurrency forecasting, balancing accuracy with computational efficiency.

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

Multiheaded Neural Network Optimization Handled by Modified Metaheuristic Optimizer for Increased Forecasting Performance

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Marko Mihajlovic,
  • Luka Jovanovic,
  • Milos Antonijevic,
  • Svetlana Andjelic,
  • Tamara Zivkovic,
  • Miodrag Zivkovic,
  • Nebojsa Bacanin

摘要

Cryptocurrencies, particularly Ethereum (ETH), have significantly influenced the global economy, offering numerous investment opportunities. However, their inherent market volatility presents challenges for accurate forecasting. This study investigates the application of multiheaded artificial neural networks (ANNs) for predicting ETH prices. To improve forecasting performance, this work introduces a modified metaheuristic optimization algorithm, a variant of the particle swarm optimization (PSO) algorithm. A comparative analysis is conducted using a publicly available dataset to evaluate the model’s effectiveness. The modified algorithm consistently delivers superior results, optimizing the multiheaded ANN model to achieve a mean squared error (MSE) of 0.43908. Collected outcomes indicate that the proposed model is a viable solution for real-world cryptocurrency forecasting, balancing accuracy with computational efficiency.