FAILURE PREDICTION IN LOGISTICS PROCESSES: DEVELOPMENT AND VALIDATION OF A PREDICTIVE MODEL

Abstract

In the current emergency configuration, operational failures in dispatch lead to delays in material delivery, compromising customer service. This work explores a proposal for the development of a machine learning model with a predictive approach, aimed at forecasting failures in logistical processes, specifically in emergency dispatches of surgical materials. The Design Science Research methodology was employed, resulting in a predictive artifact supported by the Linear Regression technique, applied through the Knime tool. For model construction, data used for training and validation were collected from the warehouse system, totaling 20,368 dispatches, of which 1,697 were for emergencies. The developed model was statistically validated with an R² of 97%, confirming the model's accuracy in predicting failures before they occur. It is concluded, therefore, that the application of predictive models in logistical processes can increase emergency response rates by reducing delays, in addition to providing valuable insights to managers.