[Paper Review] Enhancing Supply Chain Resilience: A Machine Learning Approach for Predicting Product Availability Dates Under Disruption

Mustafa Can Camur, Sandipp Krishnan Ravi+1 more|arXiv (Cornell University)|Apr 28, 2023

Supply Chain Resilience and Risk ManagementBusiness, Management and Accounting3 citations

TL;DR

This paper proposes a machine learning approach using tree-based ensemble models—Random Forest and Gradient Boosting Machine—to predict product availability dates for inbound logistics at GE Gas Power, significantly improving forecasting accuracy over traditional regression methods. The models outperformed alternatives by capturing non-linear supplier behavior and feature interactions, with feature importance highlighting 'Latest Promise Date' and 'Supplier Code' as critical predictors.

ABSTRACT

The COVID 19 pandemic and ongoing political and regional conflicts have a highly detrimental impact on the global supply chain, causing significant delays in logistics operations and international shipments. One of the most pressing concerns is the uncertainty surrounding the availability dates of products, which is critical information for companies to generate effective logistics and shipment plans. Therefore, accurately predicting availability dates plays a pivotal role in executing successful logistics operations, ultimately minimizing total transportation and inventory costs. We investigate the prediction of product availability dates for General Electric (GE) Gas Power's inbound shipments for gas and steam turbine service and manufacturing operations, utilizing both numerical and categorical features. We evaluate several regression models, including Simple Regression, Lasso Regression, Ridge Regression, Elastic Net, Random Forest (RF), Gradient Boosting Machine (GBM), and Neural Network models. Based on real world data, our experiments demonstrate that the tree based algorithms (i.e., RF and GBM) provide the best generalization error and outperforms all other regression models tested. We anticipate that our prediction models will assist companies in managing supply chain disruptions and reducing supply chain risks on a broader scale.

Motivation & Objective

• To address the critical challenge of unpredictable product availability dates in GE Gas Power's inbound supply chain, which hampers logistics and inventory planning.
• To develop a machine learning system that predicts supplier availability dates up to months in advance, enabling better shipment scheduling and cost management.
• To evaluate and compare multiple regression models, including tree-based ensembles, to identify the most accurate approach for forecasting non-linear supply chain dynamics.
• To support the integration of predictions into an optimization model for annual logistics planning, minimizing transportation and inventory costs.

Proposed method

• The study employs a range of regression models: Simple, Lasso, Ridge, Elastic Net, Random Forest (RF), Gradient Boosting Machine (GBM), and Neural Networks, trained on real-world inbound logistics data from GE Gas Power.
• Features include both numerical (e.g., product cost, amount) and categorical variables (e.g., supplier code, contract delivery date), with careful handling of missing data, particularly for 'Contract Delivery Date' which had ~50% missing values.
• Model performance is evaluated using standard regression metrics: R², RMSE, and Mean Absolute Error (MAE), with results reported on test data to assess generalization.
• Feature importance is computed using the Gini impurity criterion in tree-based models, quantifying the contribution of each feature to prediction accuracy.
• The models are trained and validated on historical product order data, with a focus on capturing non-linear relationships between supplier behavior and availability dates.
• The final prediction system is designed to be integrated into an existing mixed-integer linear programming model for annual logistics and shipment optimization.

Experimental results

Research questions

• RQ1 Which machine learning regression model provides the most accurate prediction of product availability dates in a disrupted supply chain context?
• RQ2 How do tree-based ensemble models like Random Forest and Gradient Boosting Machine compare to traditional linear models in capturing non-linear patterns in supply chain data?
• RQ3 What are the most influential features in predicting product availability dates, and how do they impact model performance?
• RQ4 To what extent does missing data in key features like 'Contract Delivery Date' affect model accuracy and reliability?
• RQ5 Can the integration of predicted availability dates into an optimization model reduce total logistics and inventory costs?

Key findings

• Tree-based models, particularly Random Forest and Gradient Boosting Machine, achieved the best generalization error and outperformed all other regression models in predicting product availability dates.
• The R² score for the best-performing model (GBM) on test data was significantly higher than that of linear models, with RMSE and MAE values indicating strong predictive accuracy.
• Feature importance analysis revealed that 'Latest Promise Date', 'Latest Need By Date', and 'Supplier Code' were the most influential predictors, while 'Product Amount' and 'Product Cost' contributed meaningfully despite not being intuitively obvious.
• The 'Contract Delivery Date' feature had less than 0.1% importance due to high missing data rates (~50%), suggesting data quality issues rather than low predictive relevance.
• Regularized models like Lasso and Ridge performed poorly due to their inability to capture non-linear relationships in the data, underscoring the need for non-linear models in this context.
• The study identifies a key limitation: non-linear supplier behavior—such as batched shipping regardless of individual promised dates—limits further improvement in prediction accuracy, suggesting a need for advanced time-series or deep learning models in future work.