IMPLEMENTATION OF INTELLIGENT PRODUCT CLASSIFICATION SYSTEMS: IMPACT ON THE EFFICIENCY OF CUSTOMS ADMINISTRATION

Abstract

The article analyzes the possibilities of using intelligent systems in the field of goods classification for customs administration. The increase in international trade volumes, the increasing complexity of logistics chains and the constant evolution of the commodity nomenclature require the modernization of processes related to the identification and assignment of HS codes of foreign economic activity. The authors investigate how machine learning algorithms, in particular the naive Bayesian classifier and artificial neural networks with the Leaky ReLU activation function, can be adapted for automated classification of goods, increasing the efficiency and reliability of solutions. The key problems of the traditional manual approach are highlighted, in particular significant time costs, dependence on the qualifications of specialists, high probability of subjective errors and limited scalability. An empirical experiment was conducted in which the results of the classification of 10,000 commodity items were analyzed by three methods: manual, using naive Bayes and a neural network. Experimental data indicate a significant increase in the accuracy of automated approaches, as well as a significant reduction in the time for processing incoming information. In particular, the use of a neural network made it possible to achieve an accuracy of 94.8% with a processing time of 60 seconds, which significantly exceeds the result of manual classification. The study also highlights the advantages of using artificial intelligence algorithms in the context of strategic management of customs resources. Reducing the need to involve a large number of specialists in routine classification processes allows optimizing the staff structure, reorienting it to analytical and supervisory activities, in particular risk assessment and detection of attempts to evade customs payments. In addition, the standardization and transparency provided by intelligent systems have a positive effect on the level of trust from business and international partners. Special attention is paid to the prospects for improving intelligent classification systems. The possibilities of implementing natural language processing (NLP) for interpreting unstructured text descriptions of goods, using computer vision for automatic identification of products by visual features, as well as the development of federated learning as a mechanism for international cooperation between customs authorities without violating data confidentiality are considered. As an example, the experience of Singapore is given, where the implementation of systems based on machine learning made it possible to reduce the processing time of customs declarations by 50% and reduce the error rate to a minimum.The results obtained confirm that intelligent systems have the potential to become a key element of the digital transformation of the customs infrastructure, contributing to the integration of Ukraine into the global economic space, harmonization of procedures with EU standards, reduction of corruption risks and increase the efficiency of customs administration at the system level.