Abstract:Accurate time series forecasting is critical for a wide range of problems with temporal data. Ensemble modeling is a well-established technique for leveraging multiple predictive models to increase accuracy and robustness, as the performance of a single predictor can be highly variable due to shifts in the underlying data distribution. This paper proposes a new methodology for building robust ensembles of time series forecasting models. Our approach utilizes Adaptive Robust Optimization (ARO) to construct a linear regression ensemble in which the models' weights can adapt over time. We demonstrate the effectiveness of our method through a series of synthetic experiments and real-world applications, including air pollution management, energy consumption forecasting, and tropical cyclone intensity forecasting. Our results show that our adaptive ensembles outperform the best ensemble member in hindsight by 16-26% in root mean square error and 14-28% in conditional value at risk and improve over competitive ensemble techniques.

What problem does this paper attempt to address?

This paper attempts to solve several key problems in time - series prediction: 1. **Improving prediction accuracy**: Time - series prediction is crucial for many practical applications with a time component, such as in the fields of healthcare operations, finance, energy, climate, etc. However, due to the unique nature of time - series data, its analysis is usually challenging, especially in predicting future values. This paper aims to improve the prediction accuracy by proposing a new method to construct a robust ensemble of time - series prediction models. 2. **Enhancing the robustness of the model**: The performance of a single prediction model may be highly volatile due to changes in the underlying data distribution. Therefore, the method proposed in this paper utilizes Adaptive Robust Optimization (ARO) to construct an ensemble of linear regression models, where the weights of the models can be adjusted over time to enhance the robustness of the model. 3. **Dealing with model drift**: Model drift refers to the phenomenon where the predictive ability of a model declines due to data changes. This paper aims to reduce the impact of model drift on prediction performance by dynamically adjusting the model weights. Specifically, the main contributions of this paper include: - **Proposing a new ensemble method**: This method is based on the ARO framework and can adjust the weights of the underlying models in real - time and dynamically, thereby constructing a prediction tool that is both robust and accurate. The detailed method is described in Section 3. - **Analyzing the robustness guarantee of the method**: An equivalent form of the method is provided for easy implementation and calculation, as detailed in Section 3.4. - **Studying the influence of hyper - parameters**: The influence of different hyper - parameters on the adaptive ensemble method is comprehensively studied through synthetic experiments, as detailed in Section 4. - **Empirical verification**: The effectiveness of the proposed ensemble method is verified in multiple practical prediction applications, including wind speed prediction (for air pollution management), energy consumption prediction, and tropical cyclone intensity prediction. The results show that this method outperforms the best single model by 16 - 26% in terms of Root Mean Square Error (RMSE) and by 14 - 28% in terms of Conditional Value - at - Risk (CVaR), as detailed in Section 5. Through these contributions, this paper demonstrates the great potential of ARO in time - series prediction and provides a new perspective for applications in the field of machine learning.

Ensemble Modeling for Time Series Forecasting: an Adaptive Robust Optimization Approach

Bayesian optimization based dynamic ensemble for time series forecasting

A novel ensemble deep learning model with dynamic error correction and multi-objective ensemble pruning for time series forecasting

Reinforcement Learning Based Dynamic Model Combination for Time Series Forecasting

Multi-output Ensembles for Multi-step Forecasting

Frequency Ensemble Based on Multi-Band Checkpoint Saving Mechanism for Time Series Forecasting

Reinforcement Learning based dynamic weighing of Ensemble Models for Time Series Forecasting

Optimizing Time-Series forecasting using stacked deep learning framework with enhanced adaptive moment estimation and error correction

Deep Learning and Linear Programming for Automated Ensemble Forecasting and Interpretation

Oblique Random Forest Ensemble Via Least Square Estimation for Time Series Forecasting.

Towards Resilient Energy Forecasting: A Robust Optimization Approach

Fully Automated Correlated Time Series Forecasting in Minutes

A new approach to extended‐range multimodel forecasting: Sequential learning algorithms

Robust Multi-Dimensional Time Series Forecasting

A Novel Bayesian Ensembling Model for Wind Power Forecasting

pTSE: A Multi-model Ensemble Method for Probabilistic Time Series Forecasting

Advanced forecasting of COVID-19 epidemic: Leveraging ensemble models, advanced optimization, and decomposition techniques

An Adaptive, Data-Driven Stacking Ensemble Learning Framework for the Short-Term Forecasting of Renewable Energy Generation

Hierarchical Ensemble-Based Feature Selection for Time Series Forecasting

Ensemble of temporal Transformers for financial time series

An ensemble approach integrating LSTM and ARIMA models for enhanced financial market predictions