Multi-Target Decision Analytics for Logistics Delay Prediction and Operational Efficiency Enhancement

Abstract

The rapid growth of logistics operations has created massive, high-velocity datasets involving routing behaviour, shipment delays, operational constraints, and resource allocation signals. Traditional logistics systems, built on manual decision-making and rigid rule-based logic, can’t keep up with this fast-changing environment. They struggle with scalability, lack predictive power, and often react too slowly to disruptions, leading to inefficiencies, inaccurate delay handling, and limited operational visibility. To address these limitations, the proposed system introduces an end-to-end Classification and Regression Trees (CART) machine learning pipeline that automates data preparation, model training, evaluation, and prediction across key logistics tasks. The system leverages K-Nearest Neighbors Classifier ad (KNN), Decision Tree (DT) Classifier and Huber-based models implemented using Stochastic Gradient Descent (SGD) estimators to predict shipment delay durations, classify rerouting requirements, determine resource allocation codes, and evaluate shipment feasibility. It supports both regression and classification workflows, integrates model persistence for reuse, and offers single-input and batch prediction capabilities through an intuitive Flask-based web application. By learning patterns from historical logistics data and producing real-time predictive insights, the proposed system significantly boosts accuracy, adaptability, and responsiveness in logistics operations. This creates a smarter, more proactive decision-making environment that reduces delays, enhances operational efficiency, and strengthens overall supply chain performance.