Attention-Based Sequence-to-Sequence Model for Time Series Imputation.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Time series data frequently present challenges such as missing values, high dimensionality, and large volumes.
• Accurate imputation of missing data is crucial for effective time series analysis and modeling.

Purpose of the Study:

• To propose an attention-based sequence-to-sequence model (ASSM) for addressing missing value imputation in high-dimensional time series.
• To enhance the model's capability in handling long-range dependencies and correlations within time series data.

Main Methods:

• The proposed ASSM model utilizes a sequence-to-sequence architecture with a Bi-directional Gated Recurrent Unit (BiGRU) encoder and a Gated Recurrent Unit (GRU) decoder.
• Incorporates self-attention in the encoder for long-range dependency capture and cross-attention in the decoder to focus on relevant encoder sequences.
• Combines feature learning and data computation for robust imputation.

Main Results:

• Experimental results on four real-world datasets show that the ASSM model significantly outperforms six other missing value imputation algorithms.
• The model achieved superior performance across four different evaluation metrics, validating its effectiveness.

Conclusions:

• The attention-based sequence-to-sequence model (ASSM) is an effective solution for imputing missing values in high-dimensional time series.
• The integration of attention mechanisms enhances the model's ability to handle complex temporal patterns and correlations.