Analytics of inventory priority level for data driven logistics and supply chain management decisions

• Main Author: Chandran, Durkahpuvanesvari
• Format: Thesis
• Language: English
• Published: 2021
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/96431/1/DurkahpuvanesvariMSC2021.pdf.pdf; http://eprints.utm.my/id/eprint/96431/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:143452

The purpose of this study is to implement analytics techniques in data driven logistics and supply chain management decisions as a measure of inventory control. Descriptive analytics is applied to make an informed decision on the service level requirement of the inventory based on its priority level. The priority level of the inventory is analysed by performing integrated ABC and FSN analysis. The priority level is determined through the ABC–FSN matrix and which later is visualised through a scatter plot using Tableau to show the distributions of the inventory with different priority levels so that it is easy to give some advice to the company on the inventory that needs higher service level. High service level, 99.9% availability and the greatest attention is proposed for the inventory classified as priority 1 since those item are classified as high risk items which frequently runs out of stock due to its demand. In order to keep updated as well as to predict the future status of an item in term of its priority level, predictive analytics is crucial. Hence, supervised machine learning technique, classification algorithm is applied in this study. Synthetic minority oversampling technique (SMOTE) is required to avoid the drawback of undersampling issue. A robust and optimised analytical model, Random Forest with the specific parameters of “max_depth” = 7, “min_samples_leaf”= 6, “min_samples_split” = 2 and “max_features”= auto is built upon holdout validation method and hyperparameter tuning. The performance measure of the model in classifying the inventory priority level is evaluated. The model accuracy obtained is 98.53% with 2685 instances correctly classified and 40 instances incorrectly classified. The weighted average of the model precision, recall and F1-score has a very good score of 0.99.