Scholarlink

<p>This study addresses the critical challenge of accurate inflation and demand forecasting, a cornerstone of modern supply chain and economic decision-making. In the current global context, characterized by heightened volatility and inflationary pressures, reliable predictive models are essential for policymakers, economists, and industry practitioners. The research problem centers on the limitations of traditional statistical approaches, such as ARIMA and ETS, which often fail to capture the non-linear dynamics and abrupt fluctuations inherent in economic and demand data. To overcome these challenges, we propose a novel Hybrid forecasting framework integrating DL architectures—including CNNs, RNNs, and LSTM—with machine learning models Hybrid(LightGBM, XGBoost) and statistical baselines Hybrid(ARIMA, Prophet). The novelty lies in an adaptive ensemble weighting mechanism that dynamically adjusts model contributions in real time, ensuring robustness across volatile contexts. The methodology employed a systematic literature review (SLR) with transparent inclusion/exclusion criteria, rigorous preprocessing of the DataCo Global dataset (covering sales, inventory, and distribution), and standardized evaluation using MAE, RMSE, MAPE, and sMAPE. Inter-rater reliability for the review process achieved <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\kappa = 0.82\)</EquationSource> </InlineEquation>, confirming strong agreement. Empirical evaluation demonstrates that the proposed Hybrid reduces RMSE by <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(92.9\%\)</EquationSource> </InlineEquation> compared to ARIMA baselines and by approximately <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(70\%\)</EquationSource> </InlineEquation> compared to standalone DL models (e.g., LSTM). These results highlight the model’s superior accuracy, resilience, and practical relevance. By uniting methodological rigor with empirical validation, this review establishes a benchmark framework for inflation and demand forecasting, with significant implications for economic policy, data science, and supply chain optimization.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Demand forecasting with artificial intelligence: a novel hybrid approach for enhanced predictive accuracy

  • 期刊论文

  • 发表日期：

    2026年3月5日

查看全文
  • Oussama Zabraoui,
  • Anas Chafi,
  • Salaheddine Kammouri Alami

摘要

This study addresses the critical challenge of accurate inflation and demand forecasting, a cornerstone of modern supply chain and economic decision-making. In the current global context, characterized by heightened volatility and inflationary pressures, reliable predictive models are essential for policymakers, economists, and industry practitioners. The research problem centers on the limitations of traditional statistical approaches, such as ARIMA and ETS, which often fail to capture the non-linear dynamics and abrupt fluctuations inherent in economic and demand data. To overcome these challenges, we propose a novel Hybrid forecasting framework integrating DL architectures—including CNNs, RNNs, and LSTM—with machine learning models Hybrid(LightGBM, XGBoost) and statistical baselines Hybrid(ARIMA, Prophet). The novelty lies in an adaptive ensemble weighting mechanism that dynamically adjusts model contributions in real time, ensuring robustness across volatile contexts. The methodology employed a systematic literature review (SLR) with transparent inclusion/exclusion criteria, rigorous preprocessing of the DataCo Global dataset (covering sales, inventory, and distribution), and standardized evaluation using MAE, RMSE, MAPE, and sMAPE. Inter-rater reliability for the review process achieved \(\kappa = 0.82\) , confirming strong agreement. Empirical evaluation demonstrates that the proposed Hybrid reduces RMSE by \(92.9\%\) compared to ARIMA baselines and by approximately \(70\%\) compared to standalone DL models (e.g., LSTM). These results highlight the model’s superior accuracy, resilience, and practical relevance. By uniting methodological rigor with empirical validation, this review establishes a benchmark framework for inflation and demand forecasting, with significant implications for economic policy, data science, and supply chain optimization.