The present project proposal aims to define a methodology to automate the process of load forecast in the logistics of food and other perishable products using classification algorithms of machine learning. In last years, the increase in the use of B2C logistics and transportation is encouraging the companies to rethink deeply logistic network and, consequently, logistics management.
Logistics networks play a crucial role to ensure customer satisfaction and fulfillment but can cause a too high level of expired products, and consequently costs and wastes. To improve the b2C logistics for perishable products one key factor is the prediction of the optimal amount of product to load on every shipment, route, destination. In fact, with the forecast of the optimal amount of perishable product (e.g. food) to load, it is possible to reduce food waste and non-value-added activities, with their associated costs and with the possibility to occupying human resources in more profitable activities, so a shrewder general costs management. Moreover, predicting the optimal load allows the optimization in the use of warehouses, centralized and locally, that can be considered a limited resource. The local warehouses commonly have limited size, so the quantity to offer to the customer should be optimized; moreover, the centralized warehouses of the logistics network serve several local distribution centers and their inventory strongly affects the performance of the selling service. The present research aims to use real case data, acquired from a specific B2C scenario (e.g. high-speed railway catering service), to develop a machine learning system and to propose a methodology able to forecast the optimal amount of food to load in the logistics operations.
As seen in the previous paragraph of the project proposal, effective models for the prediction of the optimal amount of food to load for B2C logistics and transportation of perishable products are not currently available. Commonly, the research focus on the logistics of the distribution centers and the definition of the distribution system, but not on predicting the optimal quantities of perishable products. Thus, the project proposal considers to test and compare existing predicted methods, overcoming specific issues of the B2C logistics where the expiration date is an optimization variable, with a strong focus on the definition and validation of specific logistics features, as reliable and effective forecast predictors. Moreover, there is not a general standard approach for what concerns the optimization of the prediction. A specific B2C service will be considered as a real case scenario, e.g. a catering service for high-speed railway. This application will introduce some important variabilities to cope with, as different routing needs, very variable demand patterns, warehouse boundaries in the space of available warehouses and other significative complexities.
Likewise, there is not a specific evaluation metric for to the analysis and the consideration about the results, so one of the objectives of the project proposal will recommend a general and tested approach of evaluation of the performance.
With the implementation of this project proposal, another purpose will be to encourage research on this field, opening a new starting point and a new cause for reflection with non-negligible room for improvement. In fact, this is an important research field that is not yet inspected at its potential. The real case application in the proposed project will add relevant improvements with tangible results in the research field of B2C logistics.