Open research questions in Time Series Analysis and Forecasting
92 gap statements mined from Time Series Analysis and Forecasting papers in our 4.5M-paper local library, which holds 1,080 papers on the topic — drawn mostly from each paper's own stated research gap, future-work, challenge and limitation notes, and its abstract. The ones listed below are a selection still marked open; each names the study that raised it, with a DOI link where the paper has one.
Choosing where to publish on Time Series Analysis and Forecasting? See the ranked Computer Science journals.
What the literature leaves open
Therefore, future research will focus on developing a correction method that enhances anomaly detection and correction performance in unlabeled real-world datasets. The proposed method is trained on clean (non-anomalous) windows, and its detection/correction performance may degrade when truly clean data are scarce or when anomaly labels are unavailable in real-world deployments.
A Deep Learning Approach for Time Series Data Correction:Integrating Autoencoder-Based GANs andCorrelation Analysis · 2026 · DOIDespite advances in univariate time-series (UTS) clustering, multivariate (MTS) clustering remains underexplored—multiple channels complicate inter-channel dependency modeling and amplify the accuracy–runtime trade-off.
Existing time series libraries lack coverage for generative models. There is a need for a comprehensive benchmark library for generative time series models.
GenTS: A Comprehensive Benchmark Library for Generative Time Series Models · 2026The paper does not provide a comprehensive comparison with other threshold models. The simulation experiments are limited to specific scenarios. The real-data applications are not extensively discussed.
Wavelet Based Time Series Models with Time-Varying Thresholds · 2026Most approaches focus primarily on predicting point estimates, while modeling uncertainty remains comparatively underexplored. Existing methods rarely capture the spatial interaction of residuals.
Improving Spatio-Temporal Residual Error Propagation by Mitigating Over-Squashing · 2026Building upon these findings, future lines of research for the evolution of KairosHope will focus on the following areas: 1. Geometric Invariance: Developing specific feature extraction modules for contour-type data that can be integrated into the hybrid head without compromising performance on dynamic series.
KairosHope: A Next-Generation Time-Series Foundation Model for Specialized Classification via Dual-Memory Architecture · 2026Future work will consider extending the framework to state and parameter estimation problems, multivariate and non-Gaussian noise models, and hybrid model-based deep learning architectures that combine the complementary strengths of both approaches.
An Objective Performance Evaluation of the LSTM Networks in Time Series Classification · 2026Future research can apply the framework to other application domains, such as finance and social networks. Future research can explore the use of other machine learning techniques for the classification task. Future research can investigate the use of the framework for real-time analysis of event sequences.
Analysis framework for higher-order temporal correlations with applications to human heartbeats · 2026 · DOIThe paper identifies a need for a novel analysis framework that can reveal the hierarchical structure of bursts.
Analysis framework for higher-order temporal correlations with applications to human heartbeats · 2026 · DOIThe Prophet model has limitations in automated component identification. The model has limited statistical decomposition transparency. The model assumes relatively smooth structural changes and may underperform in highly volatile empirical data.
A COMPREHENSIVE STUDY OF MODIFIED PROPHET MODEL FOR TIME SERIES COMPONENT IDENTIFICATION · 2026 · DOIThere is a need for methods that can provide explanations for time series data. Current methods have limitations, such as uncertainties in the models.
The permutation tests have limited power in short timeseries with substantial measurement error. The tests are computationally expensive for large datasets. The application of the tests to empirical datasets is limited by the availability of high-quality data.
Model-free inference of evolution from allele frequency timeseries using permutation tests · 2026 · DOIThe lack of a comprehensive time-series feature set in native Python. The proprietary implementation of hctsa is a barrier to open science and industry applications.
Future research will explore the integration of dynamic, machine-learning-driven targets to adaptively adjust bshape across alternative spatial contexts, such as solar generation networks and complex supply chain hierarchies. Future work could explore Adaptive SP-MinT, where λt varies dynamically over time based on forecasted volatility.
Shape-preserving minimum trace (SP-MinT): a regularized forecast reconciliation method for hierarchical time series · 2026 · DOIThe experimental results confirm the effectiveness of Multi-SpaCE in diverse datasets, consistently achieving good rankings across all metrics and addressing important limitations of previous approaches, such as the use of fixed-length subsequences and limited multivariate support.
Multi-SpaCE: Multi-objective subsequence-based sparse counterfactual explanations for multivariate time series classification · 2026 · DOITraditional relational database management systems are poorly suited for time series data. Feature redundancy in time series data can reduce the effectiveness of machine learning models and increase storage costs.
Problems and Approaches to Storing and Analyzing Time Series under Conditions of Feature Redundancy · 2026 · DOITo explore the use of other models and techniques, such as machine learning algorithms, to improve forecasting accuracy. To investigate the use of the methodology in real-time forecasting applications.
Multivariate reconciliation for hierarchical time series · 2026Traditional machine learning models are not able to utilize the sequential ordering of data points and temporal dependencies. Manual feature engineering presents many trade-offs such as extensive development hours, subpar performance, and inflexible feature quality.
To examine the performance of the Modified Prophet Model in different data contexts. To explore the application of the Modified Prophet Model in various domains.
A COMPREHENSIVE STUDY OF MODIFIED PROPHET MODEL FOR TIME SERIES COMPONENT IDENTIFICATION · 2026 · DOITo explore the use of other correction models. To investigate the application of the method to other fields.
Further development of pyhctsa to include additional time-series analysis methods. Application of pyhctsa to various domains, such as medical diagnosis and demand forecasting.
Evaluating the proposed approach using multiple datasets. Comparing the proposed approach with other time series storage and analysis methods. Applying the proposed approach to other industrial and IoT systems.
Problems and Approaches to Storing and Analyzing Time Series under Conditions of Feature Redundancy · 2026 · DOIInvestigating the application of GenTS to other time series analysis tasks. Developing new generative time series models and evaluating them using GenTS.
GenTS: A Comprehensive Benchmark Library for Generative Time Series Models · 2026To extend the proposed approach to other classes of threshold models. To apply the model to various fields, such as finance and economics. To compare the proposed model with other threshold models.
Wavelet Based Time Series Models with Time-Varying Thresholds · 2026The development of more powerful permutation tests for short timeseries data. The application of the tests to larger and more diverse datasets. The integration of the tests with other analytical tools to improve the interpretation of allele frequency timeseries data.
Model-free inference of evolution from allele frequency timeseries using permutation tests · 2026 · DOI
Most-cited papers in Time Series Analysis and Forecasting
- Temporal Fusion Transformers for interpretable multi-horizon time series forecasting · International Journal of Forecasting · 2021 · 2,316 citations
- Time-series forecasting with deep learning: a survey · Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences · 2021 · 1,585 citations
- The Band Pass Filter · International Economic Review · 1999 · 1,164 citations
- Why You Should Never Use the Hodrick-Prescott Filter · The Review of Economics and Statistics · 2017 · 1,098 citations
- The Gaussian Graphical Model in Cross-Sectional and Time-Series Data · Multivariate Behavioral Research · 2018 · 922 citations
- An empirical survey of data augmentation for time series classification with neural networks · PLoS ONE · 2021 · 660 citations
- SAITS: Self-attention-based imputation for time series · Expert Systems with Applications · 2023 · 532 citations
- Mel Frequency Cepstral Coefficient and its Applications: A Review · IEEE Access · 2022 · 496 citations
- A Review of Deep Learning Models for Time Series Prediction · IEEE Sensors Journal · 2019 · 468 citations
- A review of irregular time series data handling with gated recurrent neural networks · Neurocomputing · 2021 · 422 citations
Most recent work
- Deep Time Series Models: A Comprehensive Survey and Benchmark · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2026
- Time-alignment-free impact load localisation via dynamic time warping · International Journal of Mechanical Sciences · 2026
- Tree-based vs. deep learning in time series: A prescriptive framework for performance-driven model selection · Applied Soft Computing · 2026
- Out-of-Distribution Generalization in Time Series: A Survey · Information Fusion · 2026
- Cross-domain time-frequency Mamba: A more effective model for long-term time series forecasting · Knowledge-Based Systems · 2026
- Knowledge Aware State-Space Capsule Network for Multivariate Time Series Classification · IEEE Transactions on Knowledge and Data Engineering · 2026
- MambaTS: Improved selective state space models for long-term time series forecasting · Pattern Recognition · 2026
- TF-LLM: Enhanced time series analysis with time-frequency large language models · Neural Networks · 2026
- PatchFusionMLP: A scalable multi-resolution MLP framework for time series prediction · Pattern Recognition · 2026
- Bridging Distribution Gaps in Time Series Foundation Model Pretraining With Prototype-Guided Normalization · IEEE Transactions on Neural Networks and Learning Systems · 2026
Find a gap in your own Time Series Analysis and Forecasting sub-topic
This page shows what the Time Series Analysis and Forecasting literature already flags as unresolved. To narrow it to your specific question, search the Research Gap Finder: the search is free with a free account and lists the papers closest to your topic first. Unlocking that topic (50 credits, charged once) fills the comparison table from our 4.5M-paper local library and writes the gaps from its rows.
Open the Research Gap Finder →