OptiShip: Predictive Freight Cost Modeling Incorporating Regional Fuel Variability

Abstract

Properly estimating the cost of freight transportation has always been a challenge to the logistics sector, especially when considered in a region like India, where some factors are dynamic, such as the price of diesel, which keeps on fluctuating and the variable nature of delivery costs to be accounted for. The traditional estimation methods tend to be based on either contractual or stationary models, which do not capture spatial and time eras of fuel prices, and as such, they cannot cope with reality in the real world. The proposed study presents OptiShip, machine learning-based infrastructure that could provide realistic and situationally aware freight cost forecasts. The major goal is to improve the accuracy of cost estimation, which involves incorporating regional avenues of fuel price variability alongside the fundamental logistical components, namely distance and delivery time. The framework is based on a five-phase approach, which includes the data gathering, preprocessing, and pre-alignment of the diesel prices based on the spatial locations in terms of a KDTree model, training and analyzing of the models. To model the non-linear, multidimensional components of the relationship between the input features, three algorithms of ensemble learning, Random Forest, Gradient Boosting, and XGBoost are used. The hyperparameter tuning with the help of GridSearchCV and the evaluation of the performance of the models are performed through R2, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) scores. The Random Forest Regressor is found to be the best according to experimental results, having an R2 value of 0.97, an RMSE of 12.69 and an MAE of 4.94 which proves that it can model the actual representations of the highly accurate real-world logistic situation. The results indicate that regional economic indicators can be very useful to integrate into cost forecasting models and indicate that OptiShip could also be employed in real-time logistics platforms.