Leveraging Machine Learning and Genetic Algorithms for Efficient Inventory Segmentation
Deepak Kumar K, Senthil Pandi S, P. Rajesh Kumar · 2024
Particularly, inventory classification and supply chain segmentation are seen as competitive benefits in several industries. Simple rules-of-thumb like the CEF-PQR analysis are often employed in practice to classify SKUs based on fundamental principles for ranking, even if they may lead to less-than-ideal outcomes and higher expenses. To classify SKUs into replenishment policy classes that specify an inventory control rule, a demand distribution, and a group service level, we therefore offer a multi-dimensional inventory classification approach, cost-based. We also provide a classification extension that complies with a general service limitation. Our approach is to relies on ML (machine learning) classifiers, and we employ a genetic algorithm to train cost- minimization decision trees, which facilitates easy comprehension and replication of classification judgments. The emphasis on cost and operations, the simplicity of use, and the understandable are our main contributions to the literature on inventory classification. Using three industry data sets, we evaluate the method and show that the trees result in an AVG and increase of just 1.02% (3.40% with an overall service restriction) relative to the cost-optimal, where no a tree structure is required. Data that has not yet been included in the sample can be classified after trees are constructed, with an average cost increase of 1.83% (7.65%) over the best classification cost.