O uso de algoritmos de classificação para determinar estoques de segurança

Abstract

This work analyzes the application of different Artificial Neural Network algorithms with the objective of classifying candidate products for safety stock parameterization based on the purchase history of these products. The study brings discussions about the importance of the quality of the treated data and the importance of analyzing the problem, the objectives and the balance between the level of customer service, that is, the expected execution time, and the cost of inventory. K-means, affinity propagation, mean displacement and hierarchical clustering algorithms, which include the Ward and complete linkage methods, were studied. Through the application of these algorithms, it was possible to generate distribution graphs of classified points and calculate the correlation of the groups found by the different algorithms. For the presented problem, an optimal number of groups value was determined for a better classification of the products. The algorithms that showed the highest correlation were the K-means and Ward algorithms, which provided an efficient classification of data into 3 groups.