**TSI-Bench: Benchmarking Time Series Imputation** Effective imputation is crucial for time series analysis, but the lack of standardized and comprehensive benchmark platforms hinders the evaluation of imputation performance across different settings. To address this gap, the authors introduce TSI-Bench, a comprehensive benchmark suite for time series imputation using deep learning techniques. TSI-Bench standardizes experimental settings to enable fair evaluation of imputation algorithms and provides insights into the impact of domain-appropriate missingness ratios and patterns on model performance. Additionally, it offers a systematic paradigm to tailor time series forecasting algorithms for imputation purposes. **Key Contributions:** 1. **Comprehensive Benchmark:** TSI-Bench is the first standardized benchmark for time series imputation, encompassing 28 algorithms across four diverse domains (air quality, traffic, electricity, and healthcare). 2. **Research and Application-Driven Benchmarking:** It provides a research and application-driven perspective, enabling standardized analysis of imputation processes, including data simulation, model evaluation, and downstream task performance. 3. **Addressing Evaluation Gaps:** TSI-Bench ensures rigorous and equitable comparisons by providing an open-source ecosystem with standardized data preprocessing, flexible missingness simulation, metric utilization, and hyper-parameter tuning. **Experimental Results:** - **Data Perspective:** Different models exhibit significant performance variations across datasets and missingness rates, with no single model outperforming all others. - **Model Perspective:** Transformer/Attention-based models generally perform well, but RNN-based models show good performance with longer inference times and moderate model sizes. - **Downstream Task Perspective:** Imputation significantly improves the performance of downstream tasks such as classification, regression, and forecasting. **Conclusion:** TSI-Bench serves as a reference for future research and provides practical guidelines for real-world applications. The authors envision TSI-Bench as a long-term evolving project, committed to continuous development and integration of more advanced models and datasets.**TSI-Bench: Benchmarking Time Series Imputation** Effective imputation is crucial for time series analysis, but the lack of standardized and comprehensive benchmark platforms hinders the evaluation of imputation performance across different settings. To address this gap, the authors introduce TSI-Bench, a comprehensive benchmark suite for time series imputation using deep learning techniques. TSI-Bench standardizes experimental settings to enable fair evaluation of imputation algorithms and provides insights into the impact of domain-appropriate missingness ratios and patterns on model performance. Additionally, it offers a systematic paradigm to tailor time series forecasting algorithms for imputation purposes. **Key Contributions:** 1. **Comprehensive Benchmark:** TSI-Bench is the first standardized benchmark for time series imputation, encompassing 28 algorithms across four diverse domains (air quality, traffic, electricity, and healthcare). 2. **Research and Application-Driven Benchmarking:** It provides a research and application-driven perspective, enabling standardized analysis of imputation processes, including data simulation, model evaluation, and downstream task performance. 3. **Addressing Evaluation Gaps:** TSI-Bench ensures rigorous and equitable comparisons by providing an open-source ecosystem with standardized data preprocessing, flexible missingness simulation, metric utilization, and hyper-parameter tuning. **Experimental Results:** - **Data Perspective:** Different models exhibit significant performance variations across datasets and missingness rates, with no single model outperforming all others. - **Model Perspective:** Transformer/Attention-based models generally perform well, but RNN-based models show good performance with longer inference times and moderate model sizes. - **Downstream Task Perspective:** Imputation significantly improves the performance of downstream tasks such as classification, regression, and forecasting. **Conclusion:** TSI-Bench serves as a reference for future research and provides practical guidelines for real-world applications. The authors envision TSI-Bench as a long-term evolving project, committed to continuous development and integration of more advanced models and datasets.