Scholarlink

<p>This study proposes a hybrid model for financial time series forecasting, combining Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN), Multidimensional K-Nearest Neighbor (MKNN), and Backpropagation Neural Networks (BPNN). The original data are decomposed into Intrinsic Mode Functions (IMFs) and a residual wave using ICEEMDAN. The prediction process employs MKNN and BPNN to identify historical states similar to the current state, using these states as the training dataset for BPNN and the current state as the testing dataset. The predicted values of each IMF and the residual wave are then merged linearly to obtain the final forecast. To validate the efficacy of the proposed model, experiments were conducted on six stock indexes, which were derived from diverse market conditions. The model’s performance is evaluated using four metrics: Matthews correlation coefficient (MCC), Mean Absolute Percentage Error (MAPE), Mean Absolute Scaled Error (MASE), and Normalized Mean Squared Error (NMSE). Model confidence set tests further validate the model’s quality. Results show that the proposed model improves the prediction performance compared to benchmark models. These benchmark models are characterized by the following features: employing Empirical Mode Decomposition (EMD) or no decomposition, excluding the MKNN or BPNN algorithm, utilizing two-dimensional data, and relying on Euclidean distance for calculations. Extended experiments explore the model’s design and advantages.</p>

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

Forecasting financial time series using a hybrid model with ICEEMDAN and multidimensional KNN-BP

  • 期刊论文

  • 发表日期：

    2025年4月20日

查看全文
  • Yidi Pan,
  • Aijing Lin

摘要

This study proposes a hybrid model for financial time series forecasting, combining Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN), Multidimensional K-Nearest Neighbor (MKNN), and Backpropagation Neural Networks (BPNN). The original data are decomposed into Intrinsic Mode Functions (IMFs) and a residual wave using ICEEMDAN. The prediction process employs MKNN and BPNN to identify historical states similar to the current state, using these states as the training dataset for BPNN and the current state as the testing dataset. The predicted values of each IMF and the residual wave are then merged linearly to obtain the final forecast. To validate the efficacy of the proposed model, experiments were conducted on six stock indexes, which were derived from diverse market conditions. The model’s performance is evaluated using four metrics: Matthews correlation coefficient (MCC), Mean Absolute Percentage Error (MAPE), Mean Absolute Scaled Error (MASE), and Normalized Mean Squared Error (NMSE). Model confidence set tests further validate the model’s quality. Results show that the proposed model improves the prediction performance compared to benchmark models. These benchmark models are characterized by the following features: employing Empirical Mode Decomposition (EMD) or no decomposition, excluding the MKNN or BPNN algorithm, utilizing two-dimensional data, and relying on Euclidean distance for calculations. Extended experiments explore the model’s design and advantages.