Optimizing Neutrosophic Inventory Management: A Comparative Analysis of XGBoost and Random Forest Models

Nagarajan Deivanayagampillai^1,2, Thangavel Bhuvaneswari^3,*, Yasothei Suppiah³, Kanchana Anbalagan⁴

¹Department of Mathematics, Rajalakshmi Institute of Technology, Chennai, India

²Postdoctoral Researcher, Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka.Malaysia

³Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, 75450 Melaka. Malaysia.

⁴Department of Mathematics, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Saveetha University, Chennai, India. Chennai - 602 105

Emails: dnrmsu2002@yahoo.com; t.bhuvaneswari@mmu.edu.my; yasothei.suppiah@mmu.edu.my; kanchana.anbazhagan@gmail.com   

Abstract

Fuzzy sets and probabilistic methodologies have been integrated with forecasting but do not simultaneously capture the truth, indeterminacy, and falsity—really the crux of Neutrosophic Logic (NL). There is no literature investigating the incorporation of neutrosophic numbers into deep architectures, in particular into Neutrosophic Neural Networks (NNNs) for demand forecasting. This contribution fills the gap with the presentation of a Neutrosophic Neural Network (NNN) model with uncertainty explicitly included, enhancing the reliability and explain ability of demand forecasting. Deep learning-based demand forecasting strategies involving the use of Random Forest regression and XGBoosting algorithms generally do not deal with uncertainty and imprecision related with real-world demand data. The current work introduces a new model Neutrosophic Neural Network (NNN) where Neutrosophic Logic (NL) is integrated into deep learning demand forecasting. A novel neutrosophic activation function and a Neutrosophic Mean Squared Error (NMSE) loss function are proposed study, is implemented with the Random Forest regression and XGBoosting algorithms, and trained using synthetic and real-world demand data. Experimental results establish the better performance of the NNN approach about forecasting accuracy, robustness, and uncertainty handling. The sensitivity analysis also confirms the flexibility of the model with different demand patterns. The work contributes significantly towards neutrosophic deep learning and the possibility of robust and interpretable demand forecasting for supply chain and business intelligence.

Keywords: Demand Forecasting; Deep Learning; Neutrosophic Neural Network; Uncertainty Modeling; LSTM; Supply Chain Optimization