Home / Current Issue / Paper 1722628
Predictive Forecasting for Operational Resource Allocation: A Comparative Review
Subject area: Management and Commerce · Area of research: Operational Resource Forecasting
Abstract
Operational resource allocation (deciding how much staff, inventory, capacity, or service capability to position, when, and where) depends on the quality of the forecasts that feed it. This review examines the principal classes of predictive forecasting methods that inform such decisions and compares them along dimensions that matter to practitioners: predictive accuracy, interpretability, data requirements, computational and maintenance cost, and operational fit. We organise the literature into three families: classical statistical methods (exponential smoothing and Box–Jenkins ARIMA models), machine-learning methods (regression trees and their ensembles, gradient boosting, and artificial neural networks), and hybrid or combined approaches that blend the two. Drawing on large-scale forecasting competitions and comparative studies, we find that no single family dominates across all conditions. Statistical methods remain strong, robust, and economical baselines for the short, noisy, and numerous series typical of operations, while machine-learning methods offer advantages when nonlinearities, exogenous drivers, and abundant data are present. Hybrid and combined forecasts frequently reduce error and risk relative to their constituents. We connect these findings to the resource-allocation problem, arguing that accuracy alone is an incomplete criterion: forecast horizon, hierarchical structure, error asymmetry, and the cost of being wrong must shape method selection. We identify evidence gaps, especially the scarcity of controlled comparisons on operational data and inconsistent accuracy reporting, and propose directions for research and practice. The review targets analysts and managers seeking a structured, evidence-based basis for choosing forecasting methods in operational settings.
Keywords
predictive forecasting; resource allocation; time-series forecasting; machine learning; exponential smoothing; ARIMA; forecast combination; operations management
How to cite this paper
@article{1722628,
author = {Uchechi Mary-Linda Unamma, Funmilayo Ashore-Onisemo, Uzoamaka Iwuanyanwu, Ifeanyichukwu Jeffrey Okwesa},
title = {Predictive Forecasting for Operational Resource Allocation: A Comparative Review},
journal = {Iconic Research And Engineering Journals},
year = {2018},
volume = {1},
number = {9},
pages = {493-508},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1722628.pdf},
abstract = {Operational resource allocation (deciding how much staff, inventory, capacity, or service capability to position, when, and where) depends on the quality of the forecasts that feed it. This review examines the principal classes of predictive forecasting methods that inform such decisions and compares them along dimensions that matter to practitioners: predictive accuracy, interpretability, data requirements, computational and maintenance cost, and operational fit. We organise the literature into three families: classical statistical methods (exponential smoothing and Box–Jenkins ARIMA models), machine-learning methods (regression trees and their ensembles, gradient boosting, and artificial neural networks), and hybrid or combined approaches that blend the two. Drawing on large-scale forecasting competitions and comparative studies, we find that no single family dominates across all conditions. Statistical methods remain strong, robust, and economical baselines for the short, noisy, and numerous series typical of operations, while machine-learning methods offer advantages when nonlinearities, exogenous drivers, and abundant data are present. Hybrid and combined forecasts frequently reduce error and risk relative to their constituents. We connect these findings to the resource-allocation problem, arguing that accuracy alone is an incomplete criterion: forecast horizon, hierarchical structure, error asymmetry, and the cost of being wrong must shape method selection. We identify evidence gaps, especially the scarcity of controlled comparisons on operational data and inconsistent accuracy reporting, and propose directions for research and practice. The review targets analysts and managers seeking a structured, evidence-based basis for choosing forecasting methods in operational settings.},
keywords = {predictive forecasting; resource allocation; time-series forecasting; machine learning; exponential smoothing; ARIMA; forecast combination; operations management},
month = {March},
}