Global trade relies on the efficiency of port operations and logistics; prompt vessel berthing is, therefore, necessary to lower delays and related expenses. Berthing timings can be influenced by elements including ship size berth capacity. Thus, improving port operations and resource allocation depends on precisely forecasting turnaround and waiting times. This paper addresses the creation of two machine learning models such as turnaround time and waiting time models by using Huber regression, Tweedie regression, and Gradient Boosted Regression (GBR) to anticipate operational metrics in port and logistics. The datasets for both models were split 80% for training and 20% for testing therefore guaranteeing strong evaluation. Although Tweedie regression presents a flexible approach by allowing different distributions including those with non-constant variance the Huber regression model shows strong resistance to outliers, hence controlling the variations in turnaround times. GBR uses ensemble learning to aggregate several weak learners, hence raising predicted accuracy. Regarding generalization and accuracy, performance tests show that the Tweedie and GBR models exceeded the Huber model. Particularly the Tweedie model obtained a training R2 of 0.9999 and a test R2 of 0.9671; GBR demonstrated a training R2 of 1.0. This paper shows how various machine learning methods might efficiently maximize port operations and offers insightful analysis for improving operational efficiency.