A comparison of different machine learning strategies for DIFOT prediction

Abstract

Abstract : Improving results by optimizing processes execution is one of the major companies objectives. And for many of these companies, the main point to achieve better results is the good maintance of supply chain management. Optimizing the supply chain can help to ensure the customer satisfaction which can receives their order on the scheduled date and in the right quantity. Companies are constantly studying how to improve their performance in order to achieve this ideal scenario. Many management procedures are studied and many measurement techniques are tested, all of them with the aim of reducing costs and improving customer satisfaction. A widely used way that helps organizations to measure whether they are achieving this goal is the use of a KPI called Delivery In Full, On Time (DIFOT). There are several studies that deal with logistics and delivery optimization aiming to minimize delary risks and production problems, leadning to a better performance. In addition, other studies that aim to provide demand forecast and production capacity, are also widely used. These initiatives can helps companies to achieve DIFOT. A field not so explored until now is the appliance of data science techniques to assist in the process of optimizing all the production logistics and supplies delivery. This paper proposes the use of data science tools together with artificial intelligence and machine learning techiniques to bring an innovative vision to solve problems supply chain and process. This study uses a priori data, that is, sales orders features that are obtained right from the begining of sales process, and then, and one of the objetives is, to identify the correlation between features that are related to the occurrence of DIFOT. Subsequently, these features were used in machine learning models in order to identify how predictable the occurrence of DIFOT is when a new sales order is created. The results obtained demonstrate that it is possible to define predictive models that can assist in the decision-making process of organizations to ensure improvement in the supply chain and, consequently, improving KPI DIFOT performance.