Classifying Rarely Used Spare Parts Based On SVMs

Rarely Used Spare Parts (RUSP) classification is a critical problem in stock management. RUSP are highly-important items in company operation. Their typical characteristics include high cost, low frequency, long lead time and uncertainty.In this paper, firstly, we reviewed traditional ABC classification methods and their limitations in classifying RUSP. Then we introduce Support Vector Machines (SVMs) and related machine learning theory. SVMs have the advantages such as simple structure, faster classification speed, and better generalization ability and global optimized. It is suitable for complex classification problems like evaluating RUSP.Secondly, we studied a real RUSP classification case. We proposed an RUSP classifying framework based on SVM. Then we illustrate some key technical points in the framework: we compared different multi-class coding schema; we compared different kernel functions and multi-class SVM classifiers; we also proposed an improved parameter selection method of SVM named Parameter Selection via Adaptive Pattern Search (PSAPS). Numeric experiments demonstrate its efficiency.Finally, we summarized our work and point our further research direction.