Forecasting customer churn has long been a major issue in the banking sector because the early identification of customer exit is crucial for the sustainability of banks. However, modeling customer churn is hampered by imbalanced data between classification classes, where the churn class is typically significantly smaller than the no‐churn class. In this study, we examine the performance of deep neural networks for predicting customer churn in the banking sector, while incorporating various resampling techniques to overcome the challenges posed by imbalanced datasets. In this work we propose the utilization of the APTx activation function to enhance our model’s forecasting ability. In addition, we compare the effectiveness of different combinations of activation functions, optimizers, and resampling techniques to identify configurations that yield promising results for predicting customer churn. Our results offer dual insights, enriching the existing literature in the field of hyperparameter selection, imbalanced learning, and churn prediction, while also revealing that APTx can be a promising component in the field of neural networks.
{"title":"Deep Dive Into Churn Prediction in the Banking Sector: The Challenge of Hyperparameter Selection and Imbalanced Learning","authors":"Vasileios Gkonis, Ioannis Tsakalos","doi":"10.1002/for.3194","DOIUrl":"https://doi.org/10.1002/for.3194","url":null,"abstract":"Forecasting customer churn has long been a major issue in the banking sector because the early identification of customer exit is crucial for the sustainability of banks. However, modeling customer churn is hampered by imbalanced data between classification classes, where the churn class is typically significantly smaller than the no‐churn class. In this study, we examine the performance of deep neural networks for predicting customer churn in the banking sector, while incorporating various resampling techniques to overcome the challenges posed by imbalanced datasets. In this work we propose the utilization of the APTx activation function to enhance our model’s forecasting ability. In addition, we compare the effectiveness of different combinations of activation functions, optimizers, and resampling techniques to identify configurations that yield promising results for predicting customer churn. Our results offer dual insights, enriching the existing literature in the field of hyperparameter selection, imbalanced learning, and churn prediction, while also revealing that APTx can be a promising component in the field of neural networks.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"73 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191687","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
New product demand forecasting is an important but challenging process that extends to multiple sectors. The paper reviews various forecasting models across different domains, emphasizing the unique challenges of forecasting new fashion products. The challenges are multifaceted and subject to constant change, including consumer preferences, seasonality, and the influence of social media. Understanding such difficulties enables us to provide an approach for improved and flexible prediction techniques. Machine learning techniques have the potential to address these issues and improve the accuracy of fashion product demand forecasting. Various advanced algorithms, including deep learning approaches and ensemble methods, employ large datasets and real‐time data to predict demand patterns accurately. The paper suggests valuable information to experts, researchers, and decision‐makers in the fashion industry, as it addresses the unique challenges and examines innovative solutions in new product forecasting.
{"title":"Demand Forecasting New Fashion Products: A Review Paper","authors":"Anitha S., Neelakandan R.","doi":"10.1002/for.3192","DOIUrl":"https://doi.org/10.1002/for.3192","url":null,"abstract":"New product demand forecasting is an important but challenging process that extends to multiple sectors. The paper reviews various forecasting models across different domains, emphasizing the unique challenges of forecasting new fashion products. The challenges are multifaceted and subject to constant change, including consumer preferences, seasonality, and the influence of social media. Understanding such difficulties enables us to provide an approach for improved and flexible prediction techniques. Machine learning techniques have the potential to address these issues and improve the accuracy of fashion product demand forecasting. Various advanced algorithms, including deep learning approaches and ensemble methods, employ large datasets and real‐time data to predict demand patterns accurately. The paper suggests valuable information to experts, researchers, and decision‐makers in the fashion industry, as it addresses the unique challenges and examines innovative solutions in new product forecasting.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"53 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191749","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We investigate the performance of dynamic factor model nowcasting with preselected predictors in a mixed‐frequency setting. The predictors are selected via the elastic net as it is common in the targeted predictor literature. A simulation study and an application to empirical data are used to evaluate different strategies for variable selection, the influence of tuning parameters, and to determine the optimal way to handle mixed‐frequency data. We propose a novel cross‐validation approach that connects the preselection and nowcasting step. In general, we find that preselecting provides more accurate nowcasts compared with the benchmark dynamic factor model using all variables. Our newly proposed cross‐validation method outperforms the other specifications in most cases.
{"title":"Predictor Preselection for Mixed‐Frequency Dynamic Factor Models: A Simulation Study With an Empirical Application to GDP Nowcasting","authors":"Domenic Franjic, Karsten Schweikert","doi":"10.1002/for.3193","DOIUrl":"https://doi.org/10.1002/for.3193","url":null,"abstract":"We investigate the performance of dynamic factor model nowcasting with preselected predictors in a mixed‐frequency setting. The predictors are selected via the elastic net as it is common in the targeted predictor literature. A simulation study and an application to empirical data are used to evaluate different strategies for variable selection, the influence of tuning parameters, and to determine the optimal way to handle mixed‐frequency data. We propose a novel cross‐validation approach that connects the preselection and nowcasting step. In general, we find that preselecting provides more accurate nowcasts compared with the benchmark dynamic factor model using all variables. Our newly proposed cross‐validation method outperforms the other specifications in most cases.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"43 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191686","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Andreas Karathanasopoulos, Chia Chun Lo, Mitra Sovan, Mohamed Osman, Hans‐Jörg von Mettenheim, Slim Skander
By implementing a multi‐objective optimization approach in forecasting, we introduce three optimization models grey wolf optimizer, genetic algorithm, and differential evolution algorithm combined with multilayer perceptron neural networks and support vector machines to predict electricity consumption in the UAE. The hybrid models' accuracy and efficiency were evaluated using various forecasting metrics. This study's contributions are threefold: it is the first to employ such a sophisticated hybrid approach, particularly using the recently introduced grey wolf optimizer, it compares optimization techniques with the established Pearson correlation‐based method for dimensionality reduction and it represents one of the most extensive macroeconomic forecasts in the UAE using multi‐objective heuristic hybrid optimization methods. Our findings indicate that the grey wolf optimizer significantly outperforms all other models, followed by the genetic algorithm.
{"title":"A multi‐objective optimization metaheuristic hybrid technique for forecasting the electricity consumption of the UAE: A grey wolf approach","authors":"Andreas Karathanasopoulos, Chia Chun Lo, Mitra Sovan, Mohamed Osman, Hans‐Jörg von Mettenheim, Slim Skander","doi":"10.1002/for.3187","DOIUrl":"https://doi.org/10.1002/for.3187","url":null,"abstract":"By implementing a multi‐objective optimization approach in forecasting, we introduce three optimization models grey wolf optimizer, genetic algorithm, and differential evolution algorithm combined with multilayer perceptron neural networks and support vector machines to predict electricity consumption in the UAE. The hybrid models' accuracy and efficiency were evaluated using various forecasting metrics. This study's contributions are threefold: it is the first to employ such a sophisticated hybrid approach, particularly using the recently introduced grey wolf optimizer, it compares optimization techniques with the established Pearson correlation‐based method for dimensionality reduction and it represents one of the most extensive macroeconomic forecasts in the UAE using multi‐objective heuristic hybrid optimization methods. Our findings indicate that the grey wolf optimizer significantly outperforms all other models, followed by the genetic algorithm.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"19 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-08-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191688","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Forecasting is a crucial part of supply chain management. Accurate forecasts have a strong influence on supply chain performance. Many forecasting methods have been developed and adapted in various domains and industries. However, none are perfect in all contexts due to the data's characteristics and the methods' strength. Hence, we propose a new ARIMAX‐LSTM hybrid forecasting model that integrates ARIMAX and LSTM models to improve the ability to capture different combinations of linear and nonlinear patterns in time series. Our proposed model is validated in a case study of coffee demand in Vietnam. The case study results show that our proposed model outperforms the well‐known single and current hybrid models regarding performance measures and degree of association. Moreover, to prove the model's robustness, we test and compare our proposed model to the previous study for Thailand's agricultural products (pineapple, corn, and cassava). Computational results demonstrate that our hybrid model is superior in the majority of experiments. It has a strong capability of predicting complex time series data. Furthermore, our proposed method increases forecasting accuracy and enhances supply chain performance (measured by the bullwhip effect; net‐stock amplification, and transportation cost.
{"title":"Toward a smart forecasting model in supply chain management: A case study of coffee in Vietnam","authors":"Thi Thuy Hanh Nguyen, Abdelghani Bekrar, Thi Muoi Le, Mourad Abed, Anirut Kantasa‐ard","doi":"10.1002/for.3189","DOIUrl":"https://doi.org/10.1002/for.3189","url":null,"abstract":"Forecasting is a crucial part of supply chain management. Accurate forecasts have a strong influence on supply chain performance. Many forecasting methods have been developed and adapted in various domains and industries. However, none are perfect in all contexts due to the data's characteristics and the methods' strength. Hence, we propose a new ARIMAX‐LSTM hybrid forecasting model that integrates ARIMAX and LSTM models to improve the ability to capture different combinations of linear and nonlinear patterns in time series. Our proposed model is validated in a case study of coffee demand in Vietnam. The case study results show that our proposed model outperforms the well‐known single and current hybrid models regarding performance measures and degree of association. Moreover, to prove the model's robustness, we test and compare our proposed model to the previous study for Thailand's agricultural products (pineapple, corn, and cassava). Computational results demonstrate that our hybrid model is superior in the majority of experiments. It has a strong capability of predicting complex time series data. Furthermore, our proposed method increases forecasting accuracy and enhances supply chain performance (measured by the bullwhip effect; net‐stock amplification, and transportation cost.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"28 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191689","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Because the exchange rate is essentially a dynamic and nonlinear system, exchange rate forecasting has been one of the most challenging topics in the financial field. This paper proposes a novel idea of “decomposition‐reconstruction‐integration” to predict exchange rate. First, based on ICEEMDAN, the original sequences are decomposed into multifrequency IMFs. Second, we use t‐test to determine the high‐frequency IMFs, low‐frequency IMFs, and trend sequence and reconstruct the high‐frequency IMFs into a new component sequence. Third, we use CNN‐LSTM model to predict these components separately and finally get the final prediction result by integration. This paper takes the USD/RMB exchange rate as research object, and the experimental results show that (1) the fluctuations of USD/RMB exchange rate are mainly affected by the trend sequence and low‐frequency IMFs and are less affected by high‐frequency IMFs. (2) The evaluation criterions RMSE, MAE, and MAPE of the ICEEMDAN‐CNN‐LSTM model are relatively small, with values of 0.0156, 0.0112, and 0.1679, respectively, indicating that the predictive performance of the model is optimal. (3) This paper has conducted various robust tests, all of which indicate that the proposed model has high prediction accuracy and stability. In summary, this paper has certain theoretical significance and application value.
由于汇率本质上是一个动态非线性系统，因此汇率预测一直是金融领域最具挑战性的课题之一。本文提出了 "分解-重构-积分 "的汇率预测新思路。首先，以 ICEEMDAN 为基础，将原始序列分解为多频 IMF。其次，利用 t 检验确定高频 IMF、低频 IMF 和趋势序列，并将高频 IMF 重构为新的分量序列。第三，使用 CNN-LSTM 模型分别预测这些分量，最后通过整合得到最终预测结果。本文以美元兑人民币汇率为研究对象，实验结果表明：（1）美元兑人民币汇率的波动主要受趋势序列和低频 IMF 的影响，受高频 IMF 的影响较小。(2）ICEEMDAN-CNN-LSTM 模型的评价标准 RMSE、MAE、MAPE 较小，分别为 0.0156、0.0112、0.1679，表明模型的预测性能最优。(3) 本文进行了各种稳健性测试，均表明所提模型具有较高的预测精度和稳定性。综上所述，本文具有一定的理论意义和应用价值。
{"title":"Forecasting USD/RMB exchange rate using the ICEEMDAN‐CNN‐LSTM model","authors":"Yun Zhou, Xuxu Zhu","doi":"10.1002/for.3190","DOIUrl":"https://doi.org/10.1002/for.3190","url":null,"abstract":"Because the exchange rate is essentially a dynamic and nonlinear system, exchange rate forecasting has been one of the most challenging topics in the financial field. This paper proposes a novel idea of “decomposition‐reconstruction‐integration” to predict exchange rate. First, based on ICEEMDAN, the original sequences are decomposed into multifrequency IMFs. Second, we use <jats:italic>t</jats:italic>‐test to determine the high‐frequency IMFs, low‐frequency IMFs, and trend sequence and reconstruct the high‐frequency IMFs into a new component sequence. Third, we use CNN‐LSTM model to predict these components separately and finally get the final prediction result by integration. This paper takes the USD/RMB exchange rate as research object, and the experimental results show that (1) the fluctuations of USD/RMB exchange rate are mainly affected by the trend sequence and low‐frequency IMFs and are less affected by high‐frequency IMFs. (2) The evaluation criterions RMSE, MAE, and MAPE of the ICEEMDAN‐CNN‐LSTM model are relatively small, with values of 0.0156, 0.0112, and 0.1679, respectively, indicating that the predictive performance of the model is optimal. (3) This paper has conducted various robust tests, all of which indicate that the proposed model has high prediction accuracy and stability. In summary, this paper has certain theoretical significance and application value.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"162 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191691","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We derive the optimal forecasts for multivariate autoregressive time series processes subject to Markov switching in regime. Optimality means that the trace of the mean square forecast error matrix is minimized by using suitable weighting observations. Then we provide neat analytic expressions for the optimal weights in terms of the matrices involved in a state space representation of the considered process. Our matrix expressions in closed form improve computational performance since they are readily programmable. Numerical simulations and an empirical application illustrate the feasibility of the proposed approach. We provide evidence that the forecasts using optimal weights increase forecast precision and are more accurate than the traditional Markov switching alternatives.
{"title":"Forecasting Markov switching vector autoregressions: Evidence from simulation and application","authors":"Maddalena Cavicchioli","doi":"10.1002/for.3180","DOIUrl":"https://doi.org/10.1002/for.3180","url":null,"abstract":"We derive the optimal forecasts for multivariate autoregressive time series processes subject to Markov switching in regime. Optimality means that the trace of the mean square forecast error matrix is minimized by using suitable weighting observations. Then we provide neat analytic expressions for the optimal weights in terms of the matrices involved in a state space representation of the considered process. Our matrix expressions in closed form improve computational performance since they are readily programmable. Numerical simulations and an empirical application illustrate the feasibility of the proposed approach. We provide evidence that the forecasts using optimal weights increase forecast precision and are more accurate than the traditional Markov switching alternatives.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"8 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191693","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jiajia Zhang, Zhifu Tao, Jinpei Liu, Xi Liu, Huayou Chen
The definition of interval‐valued time series is now a valid tool that can be used to model uncertainty with known numerical bounds. However, how to provide accurate predictions of interval‐valued time series remains an open problem. The goal of this paper is to develop a hybrid interval‐valued time series prediction model that incorporates an intuitionistic fuzzy cognitive map and a fuzzy neural network. The causal relationship and adjacency matrix among nodes of the intuitionistic fuzzy cognitive map are defined and quantified using mutual subsethhood, in which the hesitation weight is added to the connection weight among concept nodes. The approach directly constructs concept nodes and a weight matrix for automatic recognition of intuitionistic fuzzy cognitive maps from original sequence data and combines the particle swarm optimization algorithm and back propagation algorithm to run with less manual intervention. The confidence intervals of forecasted interval values are also discussed. The developed prediction model is applied to forecast interval‐valued financial time series (i.e., the Nasdaq‐100 stock index), which is composed of daily minimum price and maximum price. The feasibility and validity of the proposed developed prediction model are shown through comparisons with some existing prediction models on interval‐valued time series.
{"title":"A hybrid interval‐valued time series prediction model incorporating intuitionistic fuzzy cognitive map and fuzzy neural network","authors":"Jiajia Zhang, Zhifu Tao, Jinpei Liu, Xi Liu, Huayou Chen","doi":"10.1002/for.3181","DOIUrl":"https://doi.org/10.1002/for.3181","url":null,"abstract":"The definition of interval‐valued time series is now a valid tool that can be used to model uncertainty with known numerical bounds. However, how to provide accurate predictions of interval‐valued time series remains an open problem. The goal of this paper is to develop a hybrid interval‐valued time series prediction model that incorporates an intuitionistic fuzzy cognitive map and a fuzzy neural network. The causal relationship and adjacency matrix among nodes of the intuitionistic fuzzy cognitive map are defined and quantified using mutual subsethhood, in which the hesitation weight is added to the connection weight among concept nodes. The approach directly constructs concept nodes and a weight matrix for automatic recognition of intuitionistic fuzzy cognitive maps from original sequence data and combines the particle swarm optimization algorithm and back propagation algorithm to run with less manual intervention. The confidence intervals of forecasted interval values are also discussed. The developed prediction model is applied to forecast interval‐valued financial time series (i.e., the Nasdaq‐100 stock index), which is composed of daily minimum price and maximum price. The feasibility and validity of the proposed developed prediction model are shown through comparisons with some existing prediction models on interval‐valued time series.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"51 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191694","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Huosong Xia, Xiaoyu Hou, Justin Zuopeng Zhang, Mohammad Zoynul Abedin
Cotton, cotton yarn, and other cotton products have frequent price volatility, increasing the difficulty for industry participants to develop rational business decision plans. To support cotton textile industry decision‐makers, we apply data mining methods to extract the main influencing factors affecting cotton yarn futures prices from big data and build a probabilistic forecasting model for cotton yarn price volatility with uncertainty assessment. Based on Explainable Artificial Intelligence (XAI) and data‐driven perspectives, we use the LassoNet algorithm to extract 18 features most relevant to the target variable from the massive data and visualize the importance values of the selected features to improve the reliability. Moreover, by combining conformal forecasting (CP) with quantile regression (QR), the uncertainty measure of the point estimation results of the long and short‐term memory (LSTM) model is applied to improve the application value of the model. Finally, SHAP (SHapley Additive exPlanations) is introduced to analyze the SHAP values of the input features on the output results and to explore in depth the interaction and mechanism of action between the input features and the target variables to improve the explainability of the model. Our model provides a “big data‐forecasting model‐decision support” decision paradigm for real‐world problems.
{"title":"A new probability forecasting model for cotton yarn futures price volatility with explainable AI and big data","authors":"Huosong Xia, Xiaoyu Hou, Justin Zuopeng Zhang, Mohammad Zoynul Abedin","doi":"10.1002/for.3185","DOIUrl":"https://doi.org/10.1002/for.3185","url":null,"abstract":"Cotton, cotton yarn, and other cotton products have frequent price volatility, increasing the difficulty for industry participants to develop rational business decision plans. To support cotton textile industry decision‐makers, we apply data mining methods to extract the main influencing factors affecting cotton yarn futures prices from big data and build a probabilistic forecasting model for cotton yarn price volatility with uncertainty assessment. Based on Explainable Artificial Intelligence (XAI) and data‐driven perspectives, we use the LassoNet algorithm to extract 18 features most relevant to the target variable from the massive data and visualize the importance values of the selected features to improve the reliability. Moreover, by combining conformal forecasting (CP) with quantile regression (QR), the uncertainty measure of the point estimation results of the long and short‐term memory (LSTM) model is applied to improve the application value of the model. Finally, SHAP (SHapley Additive exPlanations) is introduced to analyze the SHAP values of the input features on the output results and to explore in depth the interaction and mechanism of action between the input features and the target variables to improve the explainability of the model. Our model provides a “big data‐forecasting model‐decision support” decision paradigm for real‐world problems.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"4 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191692","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Dohee Kim, Eunju Lee, Imam Mustafa Kamal, Hyerim Bae
Forecasting the maritime economics index, including container volume and Baltic Panamax Index, is essential for long‐term planning and decision‐making in the shipping industry. However, studies on container volume prediction are not sufficient, and the bulk freight index has highly fluctuating characteristics, which pose a challenge in long‐term prediction. This study proposes a new hybrid framework for the long‐term prediction of the maritime economics index. The framework consists of time‐series decomposition to break down a time‐series into several components (trend, seasonality, and residual), a two‐stage attention mechanism that prioritizes important variables to increase long‐term prediction accuracy and a long short‐term memory network that predicts and combines all components to derive the final predictive outcome. Extensive experiments are conducted using the container volume data, bulk freight index data, and various external variables. The proposed framework achieved a better predictive performance than existing time‐series methods, including conventional machine learning and deep learning‐based models, in the long‐term prediction of container volume and the Baltic Panamax Index. Hence, the proposed method can help in decision‐making through accurate long‐term predictions of the maritime economics index.
{"title":"Long‐term forecasting of maritime economics index using time‐series decomposition and two‐stage attention","authors":"Dohee Kim, Eunju Lee, Imam Mustafa Kamal, Hyerim Bae","doi":"10.1002/for.3176","DOIUrl":"https://doi.org/10.1002/for.3176","url":null,"abstract":"Forecasting the maritime economics index, including container volume and Baltic Panamax Index, is essential for long‐term planning and decision‐making in the shipping industry. However, studies on container volume prediction are not sufficient, and the bulk freight index has highly fluctuating characteristics, which pose a challenge in long‐term prediction. This study proposes a new hybrid framework for the long‐term prediction of the maritime economics index. The framework consists of time‐series decomposition to break down a time‐series into several components (trend, seasonality, and residual), a two‐stage attention mechanism that prioritizes important variables to increase long‐term prediction accuracy and a long short‐term memory network that predicts and combines all components to derive the final predictive outcome. Extensive experiments are conducted using the container volume data, bulk freight index data, and various external variables. The proposed framework achieved a better predictive performance than existing time‐series methods, including conventional machine learning and deep learning‐based models, in the long‐term prediction of container volume and the Baltic Panamax Index. Hence, the proposed method can help in decision‐making through accurate long‐term predictions of the maritime economics index.","PeriodicalId":47835,"journal":{"name":"Journal of Forecasting","volume":"5 1","pages":""},"PeriodicalIF":3.4,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142191690","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：