A Systematic Framework for Machine Learning-Based Robust Project Scheduling Under Uncertainty

Project-oriented organisations continue to face challenges, such as delays and cost overruns, throughout the execution of projects. These challenges arise due to the uncertainties and risks that occur throughout project execution. Classical project scheduling approaches in the field, which typically rely on deterministic information, such as fixed activity durations, are limited in their ability to accommodate real-time project uncertainties and risks. Therefore, robust project scheduling approaches have been introduced to address this issue, through the incorporation of project buffers in the schedule. However, the majority of these approaches use arbitrary buffer sizing and allocation strategies which generate rigid baseline schedules that often require frequent rescheduling, resulting in loss of productivity, schedule instability, reduced stakeholder confidence, resource conflict and wastage, and increased costs. This makes project control and coordination more difficult for project managers. This thesis addresses these challenges by developing a Machine Learning-based Robust Project Scheduling (MLRPS) framework. The framework is designed to generate project baseline schedules that are both solution robust and quality robust, thereby minimising the need for rescheduling, enhancing overall project stability, and increasing the likelihood of the project achieving its overall goals. The proposed framework in this thesis comprises of three key components. First, a ML regression model is employed to estimate project buffer sizes dynamically by predicting delays based on the project’s characteristics, offering improved flexibility in schedule management. Second, a ML risk classification model is introduced to assess the risks associated with projects at their work package level, enabling project managers to mitigate the impact on high-risk areas and activities, and thus reducing the likelihood of delays. Third, an Activity-level Buffer Sizing and Allocation (ABSA) mechanism is developed to generate buffered baseline schedules that are both stable and maintain a high probability of timely project completion. By integrating the results from both ML models into the ABSA mechanism, the framework provides a comprehensive solution for generating solution and quality robust schedules. The novelty of the framework proposed in this thesis lies in its ability to dynamically estimate the size of project buffers based on project-specific characteristics using historical data, which ultimately leads to more stable schedules with a reduced likelihood of delays, resource wastage, and rescheduling. This thesis offers a practical and systematic decision framework for project managers, which enables the generation of robust project schedules capable of tolerating uncertainties and risks encountered during project execution. This framework is tested and validated through experiments using Monte-Carlo Simulation (MCS), and the results are compared to the Critical Chain Project Management (CCPM) approach, which is a well-known benchmark approach in the robust project scheduling field.<p></p>

Author dc:creator

• Naser Musa Al Lozi (21042389)

• Machine Learning-based Robust Project Scheduling (MLRPS)

Statement dc:rights

• All rights reserved
• Open Access after 2028-01-09

Identifier

10779/exe.32948852.v1

OAI identifier oai:identifier

oai:figshare.com:article/32948852