Journal of Forecasting最新文献

Real-world systems are often formulated as constrained optimization problems. Techniques to incorporate constraints into neural networks (NN), such as neural ordinary differential equations (Neural ODEs), have been used. However, these introduce hyperparameters that require manual tuning through trial and error, raising doubts about the successful incorporation of constraints into the generated model. This paper describes in detail the two-stage training method for Neural ODEs, a simple, effective, and penalty parameter-free approach to model constrained systems. In this approach, the constrained optimization problem is rewritten as two optimization subproblems that are solved in two stages. The first stage aims at finding feasible NN parameters by minimizing a measure of constraints violation. The second stage aims to find the optimal NN parameters by minimizing the loss function while keeping inside the feasible region. We experimentally demonstrate that our method produces models that satisfy the constraints and also improves their predictive performance, thus ensuring compliance with critical system properties and also contributing to reducing data quantity requirements. Furthermore, we show that the proposed method improves the convergence to an optimal solution and improves the explainability of Neural ODE models. Our proposed two-stage training method can be used with any NN architectures.

This study proposes a novel deep auto-optimized architecture for stock price forecasting that integrates sectoral behavior with individual stock sentiment to improve predictive accuracy. Traditional stock prediction models often focus solely on individual stock behavior, overlooking the impact of broader sectoral trends. The proposed approach utilizes advanced deep learning models, including gated recurrent units (GRU), bidirectional GRU, long short-term memory (LSTM), and bidirectional LSTM, with their hybrid ensembles. These models are built using the Keras functional API and auto ML network architecture search technology. The current deep auto-optimized multimodal framework incorporates sectoral behavior, significantly improving performance metrics. This research highlights the critical role of integrating sectoral behavior in stock price prediction models.

This paper introduces a novel spread option pricing model, the nonparametric predictive inference–based copula spread option model (NPIC-SOM), designed to evaluate the interdependence of multiple underlying assets. Through empirical analysis focused on Brent-WTI spread options, a widely traded derivative, we compare the predictive performance of the NPIC-SOM against the traditional geometric Brownian motion crack spread option model (GBM-CSOM). Our findings reveal that the NPIC-SOM not only forecasts spread option prices closer to empirical values but also captures market fluctuations more accurately than the GBM-CSOM. This superiority extends across various option types, moneyness levels and delta hedge efficiency. Furthermore, the NPIC-SOM's reliance on time-varying parameters enhances prediction accuracy, particularly for extreme market scenarios. These results indicate the practicality and efficiency of the NPIC-SOM as a robust spread option pricing model, offering valuable insights for option pricing strategies in financial markets.

Fire safety is a primary concern in safeguarding lives and property. However, it is challenging to predict fire incidents and identify potential influencing factors due to limitations of data, model accuracy and interpretability. This paper proposes a novel scheme designed to enhance predictive and explainable capabilities by integrating multi-source data, adaptive machine learning methods, and Shapley additive explanation (SHAP) tools for more effective and applicable fire safety management. The scheme shows satisfactory prediction results by leveraging the data from grid-style management systems and our proposed machine learning method with dynamic time warping distance-based time series clustering, significantly outperforming the methods merely based on time series modeling. Moreover, clustered features help to clarify the main influencing risk factors and provide clearer insights for model interpretability. With global SHAP, community clusters capturing community fire event frequency, as well as historical records on fire police rescue, smoke alarms, and fire alarms, are found to be significant risk factors among all the features over the whole communities and periods via the model interpretability analysis, implying that communities where fires used to occur frequently are more likely to occur in future, which should be highly vigilant in real fire management. With local SHAP, specific risk factors that vary across communities can be identified for any single community with a given period. We demonstrate the potential of this integrated machine learning scheme in improving the prediction accuracy and risk identification applicability of fire incidents, which contributes to more effective and customized fire safety management.

Global forecasting models (GFMs) have become essential in time series prediction, as they enable cross-learning across multiple series. Although GFMs have consistently outperformed univariate approaches, their performance decreases when applied to heterogeneous time series datasets, such as those found in economic and financial applications. Clustering techniques have been used to create homogeneous time series clusters. However, the main limitations of current clustering-based GFMs are as follows: (1) employing handcrafted features instead of deep learning and (2) there is no guarantee that the resulting clusters are optimal in terms of prediction accuracy. To address these limitations, we propose a novel deep time series clustering model that jointly optimizes clustering and forecasting accuracy. The proposed method simultaneously optimizes the reconstruction, clustering, and prediction losses to ensure clusters are optimized for accurate forecasting. In addition, it employs a neighbor-aided autoencoder to capture cluster-oriented representations, leveraging neighboring time series to improve feature learning. Furthermore, we incorporate an evolutionary learning component, which iteratively refines clusters through crossover and mutation to find optimal clusters in terms of forecasting accuracy. We evaluate our proposed method on eight publicly available datasets considering various state-of-the-art forecasting benchmarks. Results indicate that across all datasets with 2620 time series, the proposed method obtains the lowest mean symmetric mean absolute percentage error (sMAPE) of 14.90, surpassing the baseline deep clustering (15.15). It exhibits enhancements of 1.28, 0.70, and 2.29 in mean sMAPE relative to DeepAR, N-BEATS, and transformer, respectively. Furthermore, it demonstrates improvements when compared to the existing clustering-based global models. The source code of the proposed clustering method is made publicly available at https://github.com/alinowshad/Evolutionary-Neighbor-Aided-Deep-Clustering-DEEPEN.