AI-POWERED FORECASTING ALGORITHMS TO OPTIMIZE LAST MILE DELIVERY

Abstract

In the increasingly competitive landscape of supply chain and e-commerce operations, last-mile delivery forecasting forms a crucial aspect in maintaining an edge over competitors. This paper introduces a novel machine-learning method designed specifically for location-based last-mile delivery forecasting aimed at reducing operational expenses. Drawing upon historical demand data, location-specific variables, and economic indicators, an array of machine learning models, including regression models, decision trees, and neural networks, are utilized to unravel intricate patterns in the data. These models are subsequently contrasted against conventional forecasting models such as ARIMA and SARIMA. To manage potential issues associated with high-dimensional data, the paper employs principal component analysis. The ultimate model is selected through rigorous hyperparameter tuning and evaluated on an independent dataset. A visual flowchart encapsulating the entire forecasting process is also provided for better comprehension. The proposed approach showcases the immense potential of machine learning in improving last-mile delivery forecasting, paving the way for reduced operational expenses and enhanced overall efficiency in supply chain and e-commerce companies.