Analysis of Warehouse Locations in the Cold Chain: A Hybrid Approach with Principal Component Analysis and Supervised Learning: Towards a More Efficient and Resilient Logistics

¹ LASTIMI laboratory, Higher School of Technology Salé, Mohammedia School of Engineers, Mohammed V University in Rabat, Morocco Avenue Prince Héritier Sidi Mohammed, B.P.: 227 Salé Médina, Morocco
² Engineering Sciences Laboratory (LSI), FPT, Faculty of Multidisciplinary Taza, Sidi Mohamed Ben Abdellah University (USMBA), Morocco

This email address is being protected from spambots. You need JavaScript enabled to view it.

Published online: 24 June 2026

Abstract

Cold chain logistics plays a critical role in ensuring the safety and quality of perishable products, particularly in rapidly growing sectors such as agri-food and pharmaceuticals. This study proposes an exploratory approach combining Principal Component Analysis (PCA) and supervised learning models to support warehouse location decisions.

The analysis is based on a dataset covering 10 logistics zones in the Casablanca region and includes key variables such as distance, transport speed, and delivery duration. Principal Component Analysis is applied to reduce dimensionality and identify the main structural patterns in the data, with the first two principal components explaining approximately 95% of the total variance.

These components are then used as input variables in supervised learning models, including Linear Regression, Support Vector Machine (SVM), and Random Forest, to predict logistics demand. Given the limited sample size, a leave-one-out cross-validation (LOOCV) approach is adopted to ensure model evaluation.

The results indicate that the Random Forest model achieves the best predictive performance (R² = 0.38), outperforming both Linear Regression (R² = 0.08) and SVM (R² = 0.15). However, the overall predictive performance remains moderate, reflecting the limited size of the dataset and the complexity of demand dynamics.

Although the findings remain exploratory, the proposed framework provides useful insights into the structuring of logistics zones and highlights the potential of combining dimensionality reduction and machine learning techniques for supporting warehouse location planning. Further research based on larger datasets and additional variables is required to enhance the robustness and practical applicability of the approach.