Page 361 - ISC PROCEEDINGS 21.4

processing grows significantly, rendering traditional rule-based “if–then” systems
inadequate. Consequently, artificial intelligence algorithms play an increasingly important
role in analyzing these datasets and generating actionable insights for supply chain
optimization [5].
2.3. Applications of artificial intelligence in import–export logistics
Within the logistics sector, supervised learning models are widely used to predict
critical variables that support operational decision-making.
2.3.1. Market demand forecasting
With sufficient data availability, supervised learning algorithms can analyze
relationships between sales performance and multiple influencing factors, including
marketing mix variables such as pricing strategies, promotional campaigns, discount
policies, and advertising activities. Additional variables such as seasonality, weather
forecasts, historical sales records, and even customer feedback collected from social
media platforms can also be incorporated into predictive models.
Through techniques such as text mining and natural language processing, AI systems
can extract valuable insights from large volumes of unstructured data, enabling more
accurate demand forecasts and supporting more efficient logistics planning.
2.3.2. Estimated time of arrival (ETA) prediction
Another important application of AI in logistics involves predicting the Estimated
Time of Arrival (ETA) of shipments. By utilizing data from digital control tower systems
that monitor transportation activities in real time, such as truck location data and
transportation schedules—logistics specialists can develop predictive models to
determine whether shipments will arrive on schedule.
These predictive models can also incorporate external variables such as traffic
congestion and weather conditions to improve forecasting accuracy.
2.3.3. Customs clearance time estimation
Based on historical data from previously processed shipments, AI systems can
estimate waiting times at customs checkpoints. These predictions can be generated by
analyzing information such as the country of origin, shipment weight, cargo dimensions,
and product classifications.
Such predictive capabilities enable logistics operators to plan transportation
schedules more effectively and reduce delays in international trade operations.
2.3.4. Equipment downtime prediction
Machine learning models can also estimate potential failure times or the remaining
useful life of equipment components in trucks, railway wagons, or logistics machinery. By
analyzing condition-monitoring data, such as vibration levels, operational intensity, or
usage frequency-AI systems can predict maintenance needs and prevent unexpected
equipment breakdowns.
2.4. Reinforcement learning applications
Unlike supervised learning, which primarily focuses on predicting outcomes,
reinforcement learning aims to identify optimal actions based on the current state of the
system while considering the long-term consequences of decisions.
Applications of reinforcement learning in logistics include:
Multi-source procurement and multimodal transportation
When companies have access to multiple supply sources, reinforcement learning
algorithms can determine the optimal quantity of goods to procure from lower-cost
international suppliers and higher-cost domestic suppliers.

360