This study introduces the development and implementation of two machine learning models aimed at enhancing the efficiency of the dispatch process in the shipping industry – traditionally a labor-intensive, manual procedure. The primary outcome of this endeavour was a predictive recommendation system, capable of suggesting the top N carriers for a particular route and estimating their respective charges, leveraging historical shipment data from a shipping company.

The dataset employed for model training encompassed approximately 5,000 historical shipments, featuring crucial parameters such as origin, destination, and shipment weight. The preprocessing phase entailed operations such as whitespace trimming, omission of cells with missing key features, and exclusion of entries involving logistic companies, amongst other techniques.

Following the evaluation of various classification and regression models, the final selection incorporated the random forest classifier and regressor architectures, attributable to their superior testing accuracies. Subsequent to hyperparameter validation, the classification model exhibited a top-5 accuracy of 50.20% – a value affected by an imbalanced dataset. In contrast, the regression model demonstrated an accuracy of 77.50%.

Further, the two models were integrated to provide a unified carrier-price prediction based on user-input data. When trialed with a more balanced dataset, the model’s top-5 accuracy elevated to 71.77%, indicating potential for improved performance with a more diverse and balanced dataset.

While the model’s performance did not entirely align with initial expectations, the scalability potential of the model – contingent on data availability – provides a promising direction for future exploration. This study thus underscores the potential of machine learning techniques in revolutionizing dispatch processes within the shipping industry.