An Explainable AI (XAI)-Based Multi-Source Data Fusion Decision Support System for Interactive Retail Sales Demand Forecasting and Managerial Scenario Planning

• Abiodun Okunola

  Ladoke Akintola University Technology

  Author

Retail demand forecasting has been transformed by machine learning and deep learning models, yet a critical gap persists: high-performing predictive models remain opaque "black boxes," and multi-source data fusion architectures rarely incorporate interpretability mechanisms that allow managers to understand, trust, and act upon model outputs in scenario planning contexts. This study addresses this gap by designing and validating an Explainable AI (XAI)-based multi-source data fusion decision support system that integrates transaction data, web behavioral signals, promotional calendars, and macroeconomic indicators into a unified forecasting pipeline. Using a design-based research methodology with retrospective retail sales data and counterfactual scenario simulations, the study compares the proposed SHAP-augmented Multi-Modal Temporal Attention Network (MM-TAN) against baseline models including XGBoost, LightGBM, LSTM, and ARIMA. Results demonstrate that the proposed framework achieves 89.4% directional forecasting accuracy, outperforming the best baseline by 8.7 percentage points, while reducing mean absolute percentage error (MAPE) to 6.8%. SHAP-based feature attribution reveals that promotional intensity, web traffic velocity, and competitor pricing constitute the strongest demand drivers. The system enables interactive managerial scenario planning through counterfactual explanations that translate forecast variations into actionable inventory and pricing adjustments. The study contributes a replicable framework for transparent, multi-source retail forecasting and demonstrates that interpretability can be integrated into high-performance predictive pipelines without sacrificing accuracy.

• pdf

Copyright (c) 2026 Abiodun Okunola (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.