Journal article · 2025
Exploring project manager commitment using machine learning on fuzzy big data
International Journal of Project Organisation and Management, 17(2), pp. 135–152 · Published
Summary
What question does this paper answer?
Is a project manager’s organisational commitment (measured as tenure with the same employer) in successful projects related to other project attributes, and which attributes predict that commitment tenure?
What did the study find?
Analysing 439 successful US government (military) IT projects from 2015–2023 with 1,230 features, the authors found that linear regression was the best of three machine learning models, explaining 23.8% of the variance in project manager commitment tenure (MSE = 40.957, MAE = 5.3051), versus 10.8% for random forest and 4.7% for SVM. Line of business (relative relief 0.36), years of experience (0.29), team network collaboration type (0.28), cross-industry subject matter (0.27) and contract status (0.22) were the strongest predictors.
Why does it matter?
The study shows that secondary, unstructured big data about actual behaviour can be turned into predictive factors of project manager commitment, bypassing speculative survey-based perceptions. The authors argue this offers organisational decision-makers robust, data-driven strategies to improve talent retention and project success rates.
This study is part of the evidence overview What predicts IT project failure and success? Evidence on project managers, bias and team transparency from machine learning and experiments
Key findings
- The study analysed a final sample of 439 successful US government IT projects, drawn from 1,230 features collected on over 500 projects during 2015–2023, using project manager tenure with the same employer as a proxy for organisational commitment.
- Average project manager commitment tenure in the sample was 11.08 years (median 11), ranging from 1 to 25 years; the average project team size was 24 (SD = 11.4) and the mean final project budget was $29,919,153.
- Among the machine learning regression models, linear regression performed best with an R² of 0.238 (MSE = 40.957, RMSE = 6.3998, MAE = 5.3051), compared with random forest (R² = 0.108) and SVM (R² = 0.047), against a baseline R² of –0.001.
- Classification-based clustering of the ten best features produced weak results, with silhouette scores ranging only from 0.18 to 0.2, which justified moving to machine learning regression.
- The linear regression model identified 18 features as significant indicators (p < 0.05) of project manager commitment tenure, led by line of business (relative relief coefficient 0.36), experience (0.29), network collaboration type (0.28), cross-industry project subject matter (0.27) and contract status (0.22).
- Scatter-plot analysis showed more Jira/kanban collaboration in US Air Force projects where project managers had lower commitment tenure, while other lines of business collaborated more through SharePoint document sharing than virtual communications or email.
- The authors conclude that project manager commitment tenure in successful projects is related to line of business, years of experience, team network collaboration type, cross-industry subject matter and contract versus on-staff status, with about a 25% effect size in the best model.
Source: Strang & Vajjhala (2025), International Journal of Project Organisation and Management, 17(2), pp. 135–152. DOI: 10.1504/IJPOM.2025.146727
Study at a glance
| Research question | RQ1: Is project managers’ organisational commitment in successful projects related to other project attributes? RQ2: Which attributes predict project manager commitment tenure? |
|---|---|
| Design | Exploratory, pragmatic machine learning study of retrospective secondary big data; 90% confidence level set for statistical tests |
| Data | Declassified US government big data repository of IT projects (mainly Army, Navy and Air Force); 1,230 features from over 500 projects in 2015–2023, cleaned to 439 successful projects |
| Methods | Radial exploratory analysis, multi-dimensional scaling, silhouette scores, linear regression, random forest and SVM regression with ten-fold cross-validation, built in R, R Studio and Python |
| Main result | Linear regression performed best (R² = 0.238, MAE = 5.3051); random forest R² = 0.108; SVM R² = 0.047; 18 features were significant predictors (p < 0.05) |
| Implication | Behavioural big data can predict project manager commitment tenure and support data-driven talent retention and project success strategies |
| Citation | Strang & Vajjhala (2025) · DOI 10.1504/IJPOM.2025.146727 |
Abstract
This study addresses two critical organisational challenges: retaining human talent and reducing high project failure rates. Our approach diverges from traditional methods by employing machine learning (ML) to analyse retrospective big data. This study’s innovation lies in utilising secondary, unstructured data to derive predictive factors of a project manager’s (PM) commitment, moving away from the speculative nature and limited impact of survey-based perceptions. We developed a new conceptual framework that focuses on actual behaviour rather than espoused theories to identify fuzzy predictors of organisational commitment. Based on three distinct ML models, our findings reveal that one model showed a notable 25% effect size, highlighting various features connected to a PM’s tenure and organisational commitment. These insights have broad implications, offering valuable global knowledge for stakeholders in projects and programs. This study underscores the significance of non-traditional data sources in understanding and predicting critical human resource metrics, opening new avenues for organisational research and decision-making.
Abstract as published in International Journal of Project Organisation and Management.
Keywords: project management; big data analysis; talent retention; project failure rates; predictive modelling; unstructured data; behavioural analysis; human resources metrics; machine learning; ML; organisational commitment
Key terms
- Organisational commitment
- An employee’s psychological attachment and loyalty to the organisation (Meyer and Allen, 1991), comprising affective, continuance and normative commitment.
- Commitment tenure
- In this study, the number of years a project manager worked for the same employer, used as a proxy for organisational commitment.
- Coefficient of determination (R²)
- The proportion of variance in the target variable explained by a model’s input features, used here as the effect size for comparing machine learning models.
Limitations
- There was no reliable way to know that 100% of the IT projects were carried out by military employees or subcontractors.
- More than one project could exist in the data for the same project manager and this could not be checked, so the data had to be treated as sampling with replacement.
- Only successful projects were examined in this exploratory phase; future studies could examine project manager commitment in failed projects.
- The MDS clusters suggest additional, unobserved features act together, and much more research would be needed to explore the collaboration-type patterns.
How to cite
Strang, K. D., & Vajjhala, N. R. (2025). Exploring project manager commitment using machine learning on fuzzy big data. International Journal of Project Organisation and Management, 17(2), 135–152. https://doi.org/10.1504/IJPOM.2025.146727
BibTeX
@article{strang2025project,
title = {Exploring project manager commitment using machine learning on fuzzy big data},
author = {Strang, Kenneth David and Vajjhala, Narasimha Rao},
journal = {International Journal of Project Organisation and Management},
volume = {17},
number = {2},
pages = {135--152},
year = {2025},
publisher = {Inderscience},
doi = {10.1504/IJPOM.2025.146727},
url = {https://doi.org/10.1504/IJPOM.2025.146727}
}Related research