Comparative Study of Machine Learning Techniques for Inventory Classification Based on Multi-Criteria Decision-Making

Comparative Study of Machine Learning Techniques for Inventory Classification Based on Multi-Criteria Decision-Making

연구 분야: Verification

학회: ICMLT '24: Proceedings of the 2024 9th International Conference on Machine Learning Technologies

Various multicriteria inventory classification methods have been developed to overcome the limitations of conventional ABC analysis. Commonly used techniques include the analytic hierarchy process (AHP) and data envelopment analysis (DEA). However, these methods are mainly focused on classifying existing items in the inventory. Furthermore, both total inventory costs and the similarity of each group should be of concern. To address the challenge of assigning groups to new, unclassified items in the warehouse, this research proposes integrating machine learning (ML) techniques with multicriteria inventory classification. The combined approach considers both similarity and total costs, thereby improving the accuracy of inventory classification for both existing and new items based on the existing groups classified using the multicriteria approach. The result has shown that among ABC analysis, DEA, and AHP; AHP outperforms in the classification of the current inventory items of the case study factory based on the minimum total inventory cost and similarity index. To achieve the highest accuracy in inventory classification, firstly discriminant analysis (DA) and artificial neural network (ANN) were identified as the most suitable machine learning (ML) techniques to be integrated. After tuning some parameters, the best adjusted ANN model was found with the highest accuracy at 97.70% of testing data and F1 at 100%, 94.74%, and 98.25% for classes A, B, and C, respectively.