Abstract:
Artificial Intelligence (AI) and Machine Learning (ML) have undoubtedly been rising technologies, and it is expected that their prevalence will only continue to increase. These technologies have changed the way of doing business in various industries, and project management is not an exemption. AI technology will strongly influence throughout its breakthroughs the future of project management in how its tasks and milestones will be delivered and controlled.
The goal of this study is to assist project managers in better-identifying their project risks at the milestone level in complex projects to optimize success rates. The process is steeped in utilizing machine learning algorithms that would accurately identify problem types and facilitate project risk analysis. The contribution of this work is two-fold: (1) we present a dataset that can serve as a benchmark for project management risk assessment in the absence of a publicly available dataset at the time of writing this thesis, and (2) we present a proof-of-concept for the applicability and use ML methods in risk assessment using this dataset.
As such, the research project starts with an overview of how AI will heavily influence the future of project management, in addition to the evolution of AI in the discipline of project management. Furthermore, the research project identifies AI potential risks and limitations. Following this, we envision a dataset that serves as a clarificatory template for risk identification via ML. The data was set up in tabular format where each data row represents a milestone associated with data variables. Subsequently, we introduced patterns into the dataset and identify problem types manually based on specific criteria. To our knowledge, there is no publicly available dataset on project management milestone/projects and their associated problem types. Therefore, the annotated dataset we created in this work serves as a benchmark for assessing risk and for future effort in this area. The dataset will be made publicly available. As a proof-of-concept, two suitable machine learning models, each utilizing a different classification algorithm such as Decision Tree (DT) and Support Vector Machine (SVM), were trained using the dataset for predicting potential problem types. Lastly, we examined both models' performance through a test set and compare them by employing confusion matrices and various associated ML performance measures.
The evaluation performance metrics outcomes proved that the DT model outperformed the SVM model for the dataset examined.