Scholarlink

This work aims to compare Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) as fitness functions in a genetic algorithm used to optimize the hyperparameters of an Echo State Network (ESN) with five probabilistic distributions. We have used financial time series data extracted from the Brazilian Stock Market. The ESN proved to be excellent for predicting time series. However, using MAE and RMSE within the genetic algorithm was ineffective since the results did not differ significantly despite the asymmetrical data distribution. Two of the five distributions stood out, Uniform and Student’s t, showing low variability both with the use of RMSE and MAE, where we graphically observed that Uniform presented better prediction using the MAE fitness.

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

Comparing MAE and RMSE as Fitness of Genetic Algorithm for Optimizing Echo State Network Hyperparameters with Different Probabilistic Distributions

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Henrique Vaz de Araújo,
  • Fabian Corrêa Cardoso,
  • Viviane Leite Dias de Mattos,
  • Eduardo Nunes Borges,
  • Giancarlo Lucca,
  • Bruno Lopes Dalmazo,
  • Rafael Alceste Berri

摘要

This work aims to compare Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) as fitness functions in a genetic algorithm used to optimize the hyperparameters of an Echo State Network (ESN) with five probabilistic distributions. We have used financial time series data extracted from the Brazilian Stock Market. The ESN proved to be excellent for predicting time series. However, using MAE and RMSE within the genetic algorithm was ineffective since the results did not differ significantly despite the asymmetrical data distribution. Two of the five distributions stood out, Uniform and Student’s t, showing low variability both with the use of RMSE and MAE, where we graphically observed that Uniform presented better prediction using the MAE fitness.