A Data Mining Model for Clustering Food Consumption Patterns

Abstract

Object clustering frequently encounters formation of artificial clusters, which compromises data quality and reduces clustering accuracy, limited data understanding, and degraded performance metrics; and high computational time. This paper addresses these limitations by proposing an optimized system for robust food consumption pattern analysis across Nigeria. The method leverages Principal Component Analysis (PCA) to mitigate the challenges, particularly single cluster formation and high dimensionality. The system utilizes a MiniBatchKMeans algorithm. Extensive evaluation of the system was conducted through a direct comparison against a baseline MiniBatchKMeans and DBSCAN, assessing performance across critical metrics including runtime, memory consumption, and internal cluster validation scores (Silhouette, Davies-Bouldin, Calinski-Harabasz). Results demonstrate that the system achieves better high-quality clustering scores than the baseline while maintaining a significant advantage in computational efficiency, with a runtime improvement of nearly 50%.