MCDFN: supply chain demand forecasting via an explainable multi-channel data fusion network model

Abstract Accurate demand forecasting is vital for optimizing supply chain management and enhancing organizational resilience. Traditional forecasting methods, relying on simple arithmetic, often fail to capture complex patterns caused by seasonal variability and special events. Although deep learning techniques have advanced, the lack of interpretable models hampers understanding and explaining predictions. We introduce the Multi-Channel Data Fusion Network (MCDFN), a novel hybrid deep learning architecture integrating multiple data modalities for superior demand forecasting. MCDFN utilizes Convolutional Neural Networks (CNNs), Long Short-Term Memory networks (LSTMs), and Gated Recurrent Units (GRUs) to extract spatial and temporal features from time series data. Comparative benchmarking against seven other deep-learning models validates MCDFN’s efficacy, showing it outperforms its counterparts across key metrics with a mean squared error (MSE) of 23.5738, root mean squared error (RMSE) of 4.8553, mean absolute error (MAE) of 3.9991, and mean absolute percentage error (MAPE) of 20.1575%. Theil’s U statistic of 0.1181 ( $$U&lt;1$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mi>U</mml:mi> <mml:mo>&lt;</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> </mml:math> ) of MCDFN indicates its superiority over the naive forecasting approach, and a 10-fold cross-validated statistical paired t -test with a p -value of 5% indicated no significant difference between MCDFN’s predictions and actual values. To address the “black box” nature of MCDFN, we employ explainable AI techniques such as ShapTime and Permutation Feature Importance, offering insights into model decision-making processes. This research advances demand forecasting methodologies and provides practical guidelines for integrating MCDFN into existing supply chain systems.