Optimizing delivery logistics: a comparative study of clustering algorithms for zoning restaurant orders

Abstract

The rise in restaurant delivery demand, increased by the Covid-19 pandemic, has motivated restaurants to partner with online food delivery platforms or develop self-logistics systems. Effective delivery zone segregation is critical for optimizing operations and enhancing customer satisfaction. This study investigates the optimal clustering algorithm for segregating delivery locations into zones using unsupervised machine learning. A synthetic dataset of 400 delivery locations across 10 zones in Manhattan, New York, was generated, with coordinates standardized for clustering. Six algorithms - K-Means, Mini-Batch K-Means, Agglomerative, BIRCH, Spectral, and Gaussian Mixture were evaluated based on internal (Silhouette Score, Davies-Bouldin Index, Calinski-Harabasz Index) and external (Rand Index, Adjusted Rand Index, Mutual Information, Adjusted Mutual Information, Normalized Mutual Information, Fowlkes-Mallows Index, Homogeneity, Completeness, V-measure) metrics. Results indicate that K-Means performs better in internal metrics, making it ideal for restaurants defining zones for the first time, while Mini-Batch K-Means tops in external metrics, suitable for validating or refining existing zones. These findings offer restaurants a data-driven approach to enhance delivery efficiency and customer service, with implications for broader logistics applications.