Scholarlink

This study explores meta-learning based on recurrent neural networks to combine forecasts generated by diverse models. While conventional forecast combination methods typically rely on basic averaging or linear regression, the application of machine learning opens the door to more advanced and efficient forecast aggregation via meta-learning. We define and evaluate different meta-learning scenarios tailored to time series with complex seasonality, leveraging the capabilities of modern recurrent neural networks to incorporate relevant information from recent time points. We specifically assess the performance of LSTM and GRU meta-learners, showcasing their superiority over conventional methods such as averaging and linear combination.

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

Meta-learning Based on Recurrent Neural Networks for Ensembling Forecasts of Time Series with Multiple Seasonal Patterns

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Grzegorz Dudek

摘要

This study explores meta-learning based on recurrent neural networks to combine forecasts generated by diverse models. While conventional forecast combination methods typically rely on basic averaging or linear regression, the application of machine learning opens the door to more advanced and efficient forecast aggregation via meta-learning. We define and evaluate different meta-learning scenarios tailored to time series with complex seasonality, leveraging the capabilities of modern recurrent neural networks to incorporate relevant information from recent time points. We specifically assess the performance of LSTM and GRU meta-learners, showcasing their superiority over conventional methods such as averaging and linear combination.