Application of bidirectional multilayer neural network for restoring missing values in time series

This scientific article investigates and analyzes the potential application of a complex architecture of bidirectional multilayer neural networks to address the issue of filling missing values in time series data. The problem of missing data is one of the key and critically significant challenges in the context of time series analysis, as the presence of missing values can substantially and negatively impact the reliability and accuracy of models, as well as the adequacy of forecasting. Research in this area highlights that ignoring missing data or using inadequate methods to fill them can lead to distortions in results, which ultimately affects the success and reliability of predictive models and analytical conclusions. Therefore, the development and application of effective methods for handling missing values is an important task to ensure high quality and accuracy in time series modeling. This study proposes a method based on the architecture of bidirectional gated recurrent units, which effectively fills in missing values by taking into account the bidirectional context of the time series. To evaluate the effectiveness of the proposed model, experiments have been conducted on various datasets, including both synthetic and real time series. The results have showed that the proposed approach outperforms traditional methods for filling missing values. The application of the bidirectional neural network model provides a powerful tool for preprocessing temporal data, opening new possibilities for further analysis and modeling of time series.