Optimization of Cargo Loading System in Logistics Delivery Services using Machine Learning at a Logistics Companies

Abstract

A more precise data-driven approach is required to optimize lead time estimation and improve service quality. This study aims to evaluate and enhance lead time accuracy by optimizing cargo loading into containers using shipment data, including item length, width, height, weight, and volume, as well as vehicle loading capacity. The data are processed to optimize the loading process using a Genetic Algorithm, combined with a Random Forest model for determining cargo stacking and rotation. The dataset is analyzed using the CRISP-DM methodology to identify patterns, trends, and inter-variable relationships that influence the optimization of cargo placement within containers. These algorithms were selected due to their ability to capture complex relational patterns and their relevance to logistics shipment data. Model performance is evaluated using accuracy metrics and a confusion matrix to comprehensively assess predictive performance. In addition, the results of the machine learning–based models are compared to identify significant improvements in estimation accuracy. The results indicate that the Genetic Algorithm achieved a fitness value of 0.836142 in Scenario 1 without Random Forest and 3.127948 in Scenario 2 when combined with Random Forest. Furthermore, the Random Forest model achieved an accuracy of 99.23% for stacking prediction and 99.33% for rotation prediction. The developed system effectively supports optimal cargo loading optimization through accurate predictive models, enabling data-driven decision-making. With the implementation of this model, logistics companies can improve operational efficiency, minimize the risk of delays, and deliver superior customer service.