A machine learning approach for forecasting hierarchical time series

Mancuso, Paolo; Piccialli, Veronica; Sudoso, Antonio M.

doi:10.1016/j.eswa.2021.115102

Cited by 41 publications

(23 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They evaluated their approach on national-scale Brazilian electrical power production as well as Australian domestic tourism data. In another work, Mancuso et al [28] proposed a method to unify the two prevailing processes that are forecasting and reconciliation. By including hierarchical information in the forecasting process through a customized loss function, they allow the network to train towards reconciled forecasts using a top-down disaggregation process.…”

Section: Reconciliation Approachesmentioning

confidence: 99%

Hierarchical learning, forecasting coherent spatio-temporal individual and aggregated building loads

Leprince¹,

Madsen²,

Møller³

et al. 2023

Preprint

View full text Add to dashboard Cite

Optimal decision-making compels us to anticipate the future at different horizons. However, in many domains connecting together predictions from multiple time horizons and abstractions levels across their organization becomes all the more important, else decision-makers would be planning using separate and possibly conflicting views of the future. This notably applies to smart grid operation. To optimally manage energy flows in such systems, accurate and coherent predictions must be made across varying aggregation levels and horizons. Such hierarchical structures are said to be coherent when values at different scales are equal when brought to the same level, else would need to be reconciled. With this work, we propose a novel multi-dimensional hierarchical forecasting method built upon structurally-informed machine-learning regressors and established hierarchical reconciliation taxonomy. A generic formulation of multi-dimensional hierarchies, reconciling spatial and temporal hierarchies under a common frame is initially defined. Next, a coherency-informed hierarchical learner is developed built upon a custom loss function leveraging optimal reconciliation methods. Coherency of the produced hierarchical forecasts is then secured using similar reconciliation technics. The outcome is a unified and coherent forecast across all examined dimensions, granting decision-makers a common view of the future serving aligned decision-making. The method is evaluated on two different case studies to predict building electrical loads across spatial, temporal, and spatio-temporal hierarchies, benchmarked against base and multi-task regressors. Although the regressor natively profits from computationally efficient learning thanks to the unification of independent forecasts into a global multi-task model, results displayed disparate performances, demonstrating the value of hierarchical-coherent learning in only one setting. Yet, supported by a comprehensive result analysis, existing obstacles were clearly delineated, presenting distinct pathways for future work. Particularly, investigating reduced number of model weights and varying native multi-output regressors can tackle the two preeminent challenges that are the curse of dimensionality and coherency-learning complications from scaled trees respectively. Overall, the paper expands and unites traditionally disjointed hierarchical forecasting methods providing a fertile route toward a novel generation of forecasting regressors.

show abstract

Section: Reconciliation Approachesmentioning

confidence: 99%

Hierarchical learning, forecasting coherent spatio-temporal individual and aggregated building loads

Leprince¹,

Madsen²,

Møller³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The accuracy of the model is assessed by Mean Square Error (MSE) and shows that all the models have good levels of training accuracy, while ELM has better testing accuracy, training speed, and testing speed. Mancuso et al have used a deep neural network approach to forecast a hierarchical time series data (Mancuso et al, 2021).…”

Section: Related Workmentioning

confidence: 99%

Trend analysis and forecasting of publication activities by Indian computer science researchers during the period of 2010–23

Kathiria

Arolkar

2022

Expert Systems

View full text Add to dashboard Cite

Huge collections of published research documents are available in various repositories in Indian universities and research organizations. The efficient retrieval, estimation of research trends, and identification of research gaps with the global trend in different research areas, can be a great guiding tool for regulating research in the appropriate direction. This research attempts to analyse the trend of research activities carried out by Indian researchers in the discipline of Computer Science from 2010 to 19 and forecast the upcoming research trend for the years 2020-23. A repository of the abstracts based on domains given in the Computer Science Ontology (CSO) published by the Indian researchers is developed from the Scopus database. Document Index Graph document representation model is used to store the repository, shared phrases across the documents are extracted and phrase-based similarity is computed. Combining the Single-term and Phrase-based similarity, the hybrid similarity is generated and similar documents are clustered using the DBSCAN clustering technique. Topics are identified for each cluster using the Latent Dirichlet Allocation algorithm and are automatically labelled using CSO. For each topic, the trend analysis and forecasting have been done using the Auto-Regressive Integrated Moving Average. For the assessment of the forecasting performance, the dataset from 2010 to 17 is used as a training dataset and 2018-19 as a testing dataset. The average forecasting for the year 2018 for all CSO domains belongs to the Good forecasting category with Mean Absolute Percentage Error (MAPE) 18.34, and 2019 shows reasonable forecasting with MAPE 30.20 as per the MAPE interpretation given by Lewis. For each topic the average forecasting for years 2018-19 shows either Highly accurate, Good or Reasonable forecasting. As a result, the top four domains for the years 2020-23 are also identified which can help initial researchers in the identification of a relevant topic for research and exploration.

show abstract

“…Nonetheless, the best forecasting method that was found to be most often the most accurate was exponential smoothing, the workhorse of many forecasting systems in practice, the performance of which can be augmented substantially when promotions and other indicators are included [11]. The usage of ML methods by retailers remains an open question, as it needs to ensure that any forecast value added is meaningful given the extra costs [4,[21][22][23]: skilled data scientists, significant amount of time for training the models, sufficient computational and data infrastructures, among other issues [24][25][26][27][28][29]. Spiliotis et al [30] used the M5 data to evaluate the forecasting and inventory performance of both established statistical approaches and advanced ML methods and concluded that simple methods may result in similar if not lower monetary costs than more sophisticated approaches.…”

Section: Introductionmentioning

confidence: 99%

Forecasting Seasonal Sales with Many Drivers: Shrinkage or Dimensionality Reduction?

Ramos

Oliveira

Kourentzes

et al. 2022

ASI

View full text Add to dashboard Cite

Retailers depend on accurate forecasts of product sales at the Store × SKU level to efficiently manage their inventory. Consequently, there has been increasing interest in identifying more advanced statistical techniques that lead to accuracy improvements. However, the inclusion of multiple drivers affecting demand into commonly used ARIMA and ETS models is not straightforward, particularly when many explanatory variables are available. Moreover, regularization regression models that shrink the model’s parameters allow for the inclusion of a lot of relevant information but do not intrinsically handle the dynamics of the demand. These problems have not been addressed by previous studies. Nevertheless, multiple simultaneous effects interacting are common in retailing. To be successful, any approach needs to be automatic, robust and efficiently scaleable. In this study, we design novel approaches to forecast retailer product sales taking into account the main drivers which affect SKU demand at store level. To address the variable selection challenge, the use of dimensionality reduction via principal components analysis (PCA) and shrinkage estimators was investigated. The empirical results, using a case study of supermarket sales in Portugal, show that both PCA and shrinkage are useful and result in gains in forecast accuracy in the order of 10% over benchmarks while offering insights on the impact of promotions. Focusing on the promotional periods, PCA-based models perform strongly, while shrinkage estimators over-shrink. For the non-promotional periods, shrinkage estimators significantly outperform the alternatives.

show abstract

A machine learning approach for forecasting hierarchical time series

Cited by 41 publications

References 32 publications

Hierarchical learning, forecasting coherent spatio-temporal individual and aggregated building loads

Hierarchical learning, forecasting coherent spatio-temporal individual and aggregated building loads

Trend analysis and forecasting of publication activities by Indian computer science researchers during the period of 2010–23

Forecasting Seasonal Sales with Many Drivers: Shrinkage or Dimensionality Reduction?

Contact Info

Product

Resources

About