Document Type : Research Paper

Authors

• Roya Soltani ¹
• Ali Nobakhti ²

¹ Assistant Professor, Department of Industrial Engineering, Faculty of Engineering, Khatam University, Tehran, Iran

² MSc in industrial Engineering, Department of Industrial Engineering, Faculty of Engineering, Khatam University, Tehran, Iran

Abstract

In this paper, a recommender system based on a multilayer feedforward artificial neural network (ANN) trained by the Levenberg–Marquardt backpropagation algorithm, optimized using a genetic algorithm (GA) to fine-tune both network structure and weights, is proposed to predict competency and recommend project managers in project-oriented organizations. The system considers both hard and soft skills, which are essential for sustainable development. The performance of the proposed system was evaluated by a case study within the Iranian construction industry, utilizing the experience of 80 senior managers and experts from the Ministry of Roads and Urban Development of Iran. The results demonstrate the high accuracy of the proposed system in identifying competent project managers. To validate the system, its performance was compared with existing methods in the literature, showing superior accuracy in terms of MSE and RMSE metrics.
Introduction
In today’s dynamic business environment, projects operate within a VUCA context—characterized by volatility, uncertainty, complexity, and ambiguity—that significantly influences managerial decision-making and project outcomes. Rapid technological advancements, economic fluctuations, the complex nature of stakeholder interactions, and resource constraints have made project management an increasingly challenging undertaking. Consequently, the competence of project managers to address a wide range of human resource, technical, and economic challenges, along with their ability to build effective communication and collaboration networks, is a crucial determinant of project success (Omoush, 2020). To ensure successful project delivery, project managers must demonstrate a sound understanding of environmental dynamics and make informed, adaptive decisions that integrate both hard and soft managerial skills—skills that are now more critical than ever for achieving sustainable development. Such abilities reflect the professional competence and strategic agility required for timely and effective decision-making (Karki & Hadikusumo, 2023).
Selecting competent project managers through a data-driven recommender system that matches the desired managerial skills can substantially enhance the effectiveness of project-based organizations. Such a system can transform the manager selection process from subjective judgments to evidence-based decision-making. This approach not only improves the precision of identifying qualified managers but also contributes to better human resource allocation, reduced managerial risk, and enhanced overall project performance. Ultimately, adopting a data-driven recommender framework enables project-based organizations to strengthen their managerial capabilities and achieve a sustainable competitive advantage.
Research Questions

How can a smart recommender system be designed—by integrating ANN and GA—to accurately identify competent project managers in project-based organizations?
To what extent can optimizing the parameters of an ANN using a GA enhance the accuracy of the recommender system?

Literature review
In the literature, various data-driven methods have been developed using machine learning approaches to enhance decision-making, resolve conflicts, and improve project performance, productivity, safety, and workflow in the field of project management. A comprehensive review of the literature reveals that existing predictive models in project management have predominantly focused on forecasting various project outcomes such as quality (Najafi Zangeneh et al., 2020; Fan, 2025), infrastructure costs (Soltanian et al., 2023; Dan, 2024; Chen, 2024; Effat, 2025; Al-Gahtani et al., 2025), dispute occurrences and litigation outcomes (Ayhan et al., 2021), delays (Awada et al., 2021), and construction crew productivity (Sadatnya et al., 2023) through the application of diverse classification algorithms (see Table 1). Despite these advances, the literature lacks studies that aim to develop a predictive model capable of accurately assessing project manager competency using a hybrid framework that combines ANN with metaheuristics. Employing such an approach could provide a robust mechanism for identifying competent project managers and, consequently, enhance the likelihood of successful project delivery in complex and dynamic construction environments.
However, in the literature, the combination of ANN with metaheuristics has been employed to improve prediction accuracy across various domains. These domains include stock market forecasting (Sharma et al., 2022), electricity consumption demand prediction (Azadeh et al., 2007), and patient mortality prediction (Dybowski et al., 1996) (see Table 2).
Methodology
The steps of the proposed recommender system for identifying competent project managers are as follows:

Data preparation: First, a database comprising data related to competency is established and quantified based on the experience of 80 senior managers and specialists. Subsequently, data cleaning is performed, and records with missing values, outliers, or inconsistencies are removed from the database. Finally, 70% of the data is randomly selected for training and 30% for testing.
Neural network architecture design: A feedforward multilayer ANN is designed based on the number of hidden layers (i.e., 1, 2, or 3) and the number of neurons per layer (i.e., 2, 4, or 8). The network is trained using the Levenberg–Marquardt algorithm. After training and testing, the optimal network structure is selected based on MSE and RMSE metrics.
Optimization of neural network weights using genetic algorithm: The weights of the designed ANN are optimized using a GA to improve the network’s predictive performance.
Training and testing the ANN-GA recommender system: The ANN optimized by the GA is first trained and then tested. The performance of the proposed ANN-GA recommender system in identifying competent project managers is evaluated based on the MSE and RMSE criteria.

Results
The ANN model with a three-hidden-layer architecture and 2 neurons per layer demonstrated the best performance in terms of MSE and RMSE, with values of 0.351 and 0.593, respectively. This indicates that the designed network effectively predicts project manager competency. To further enhance prediction accuracy, the network weights were optimized using a GA. The resulting ANN-GA recommender system achieved an MSE of 0.094 and an RMSE of 0.307, showing significantly higher accuracy in identifying competent project managers compared to the non-optimized network (MSE = 0.351, RMSE = 0.593). These findings highlight the effectiveness of combining ANN with GA for data-driven competency assessment.
To validate the proposed recommender system for identifying competent project managers, its prediction error was compared with the algorithms reported by Karki and Hadikusumo (2023). As shown in Table 12, the proposed system demonstrates superior accuracy, highlighting its effectiveness over existing methods.
Discussion
Effective project management in the construction industry, a complex and high-risk sector, requires managers capable of making informed decisions under VUCA conditions. Instead of subjective judgments, experiential biases, and unstructured evaluations, the proposed recommender system can help project-based companies use data-driven, intelligent tools to identify more competent managers, improving project productivity while significantly reducing costs associated with poor managerial decisions. Additionally, the proposed system can serve as a decision-support tool for hiring new project managers or promoting existing ones by analyzing past performance and predicting their potential success in future projects.
Conclusion
The proposed recommender system integrates an ANN with a GA to identify and select competent project managers with high accuracy. Leveraging historical data and uncovering hidden patterns, the optimized ANN accurately predicts managerial competencies based on defined criteria. Validation against existing approaches demonstrates that the GA significantly enhances predictive accuracy, highlighting the system’s potential to improve managerial selection and project outcomes in practice.

Keywords

• Construction Projects
• Recommender System
• Project Managers’ Competence
• Soft Skills
• Hard Skills
• Artificial Neural Network
• Genetic Algorithm

Main Subjects

• project management