كنفرانس بين المللی مهندسی صنايع و سيستم ها

صفحه اصلی / دوازدهمین كنفرانس بين المللی مهندسی صنايع و سيستم ها

A Data-Driven Machine Learning Framework for Activity-Level Schedule Deviation Prediction in Automotive R&D Projects

نویسندگان :

Marjan Layeghi¹ Sanaz Jamali²

1- دانشگاه اراک 2- دانشگاه امیر مبیر

کلمات کلیدی :

Machine learning،Project scheduling،Schedule deviation prediction،Automotive R&D،XGBoost،Decision support

چکیده :

Activity-level schedule deviation prediction is critical in automotive research and development (R&D) projects, where engineering uncertainty, iterative design, and interdependent activities can propagate local delays across the development program. This study develops an explainable and leakage-resistant machine-learning framework using approximately 1,300 activity-level observations from 20 completed automotive R&D projects executed between 2023 and 2025. Project identity is treated as the grouping variable throughout data partitioning, hyperparameter tuning, and validation so that activities from the same project never appear simultaneously in training and validation folds. Six regression models-Linear Regression, Decision Tree Regression, Random Forest Regression, XGBoost, MLP, and FT-Transformer-are evaluated under the same project-grouped validation protocol. Performance variability is quantified using fold-wise performance dispersion. SHAP and permutation importance are used to identify influential schedule drivers, while sensitivity and feature-rank stability analyses assess robustness. XGBoost achieved the best grouped performance, with an R² of 0.954 ± 0.016, MAE of 0.243 ± 0.031 days, and RMSE of 0.785 ± 0.150 days.