Researchers Advance Time Series Forecasting with Novel TS-RAG Approach

Researchers have proposed various methods to improve the performance of large language models (LLMs) in time series forecasting, including the use of retrieval-augmented generation (RAG) and the development of novel architectures such as TS-RAG. These methods have shown promising results in improving the accuracy of LLMs in time series forecasting tasks. However, the application of RAG in time series forecasting remains limited, and most time series models are constrained by limited training data, smaller parameter scales, and a lack of extensive generative capabilities. To address these challenges, a novel approach, TS-RAG, has been proposed, which leverages RAG to enhance forecasting performance. The framework introduces specially designed reference tokens to effectively fuse information from the input sequence with that from retrieved similar sequences, enabling a more robust capture of complex temporal dynamics. Experimental results demonstrate that TS-RAG achieves consistent state-of-the-art performance across several real-world forecasting benchmarks.

A hybrid machine learning framework has been developed for herd-level cattle growth pattern and weight gain forecasting in grazing-based production systems. The framework integrates weekly live weight observations, demographic variables, and lagged environmental predictors into structured forecasting datasets. Herd-level forecasting trajectories were generated through temporal aggregation of animal-level predictions. Four hybrid architecture families were evaluated, including residual, stacked, cascade, and ensemble-assisted frameworks. ARIMA, LSTM, and GRU models were used as comparative baselines. Independent testing demonstrated strong predictive agreement across multiple forecasting horizons. The cascade GB to RF to NN architecture achieved the best performance, with a test R^2 of 0.889, RMSE of 21.319 kg, and MAE of 15.462 kg. Hybrid architectures maintained greater robustness than recurrent sequential models under sparse observation conditions.

A controlled multimodal benchmark, C-SUITEBENCH, has been introduced to evaluate the decision-making ability of large language models in executive business decisions. The benchmark includes five decision tasks under paired text-only and multimodal conditions across 50 scenarios. The results show that multimodal inputs consistently improve evidence-centric reasoning, with the largest and most reliable gains appearing in risk forecasting and board-facing justification. However, a multimodal integration paradox was uncovered, where adding visual business information degrades constrained resource allocation for all nine models, even as visual grounding itself improves.

A benchmark revision approach has been proposed to improve the realism of synthetic clinical benchmarks under utility constraints. The approach is formulated as utility-constrained realism improvement, where dataset changes should increase realism while staying above an operational utility floor. The results show that two deterministic revisions improve the realism of the benchmark while remaining above the current utility floor, whereas a naive densification control preserves unrealistic templating.

A hybrid machine learning framework has been developed for herd-level cattle growth pattern and weight gain forecasting in grazing-based production systems. The framework integrates weekly live weight observations, demographic variables, and lagged environmental predictors into structured forecasting datasets. Herd-level forecasting trajectories were generated through temporal aggregation of animal-level predictions. Four hybrid architecture families were evaluated, including residual, stacked, cascade, and ensemble-assisted frameworks. ARIMA, LSTM, and GRU models were used as comparative baselines. Independent testing demonstrated strong predictive agreement across multiple forecasting horizons.

A novel approach, TS-RAG, has been proposed to leverage retrieval-augmented generation (RAG) to enhance forecasting performance in time series forecasting tasks. The framework introduces specially designed reference tokens to effectively fuse information from the input sequence with that from retrieved similar sequences, enabling a more robust capture of complex temporal dynamics. Experimental results demonstrate that TS-RAG achieves consistent state-of-the-art performance across several real-world forecasting benchmarks.

A hybrid machine learning framework has been developed for herd-level cattle growth pattern and weight gain forecasting in grazing-based production systems. The framework integrates weekly live weight observations, demographic variables, and lagged environmental predictors into structured forecasting datasets. Herd-level forecasting trajectories were generated through temporal aggregation of animal-level predictions. Four hybrid architecture families were evaluated, including residual, stacked, cascade, and ensemble-assisted frameworks. ARIMA, LSTM, and GRU models were used as comparative baselines.

A controlled multimodal benchmark, C-SUITEBENCH, has been introduced to evaluate the decision-making ability of large language models in executive business decisions. The benchmark includes five decision tasks under paired text-only and multimodal conditions across 50 scenarios. The results show that multimodal inputs consistently improve evidence-centric reasoning, with the largest and most reliable gains appearing in risk forecasting and board-facing justification.

A benchmark revision approach has been proposed to improve the realism of synthetic clinical benchmarks under utility constraints. The approach is formulated as utility-constrained realism improvement, where dataset changes should increase realism while staying above an operational utility floor. The results show that two deterministic revisions improve the realism of the benchmark while remaining above the current utility floor, whereas a naive densification control preserves unrealistic templating.

A novel approach, TS-RAG, has been proposed to leverage retrieval-augmented generation (RAG) to enhance forecasting performance in time series forecasting tasks. The framework introduces specially designed reference tokens to effectively fuse information from the input sequence with that from retrieved similar sequences, enabling a more robust capture of complex temporal dynamics. Experimental results demonstrate that TS-RAG achieves consistent state-of-the-art performance across several real-world forecasting benchmarks.