2026· Research on Engineering Structures and Materials· 0 citations
Abstract
Irrigation projects are essential for ensuring water security and sustainable agricultural development; however, they are highly susceptible to human, managerial, equipment, and environmental risks that contribute to cost overruns and schedule delays. Although machine learning (ML) has been widely applied to project performance prediction, its integration with validated safety indicators and explainable artificial intelligence for irrigation projects remains largely unexplored. Therefore, this study proposes an explainable machine learning framework for predicting cost overruns and schedule delays using safety indicators. A comprehensive dataset comprising 50 irrigation projects implemented in Iraq between 2022 and 2025 was developed. Twenty-three safety indicators were identified through a systematic literature review, refined through expert consultation, and validated using Partial Least Squares Structural Equation Modeling (PLS-SEM). Four ML algorithms, namely Decision Tree (DT), Random Forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost), were developed and evaluated using 5-fold cross-validation. Model performance was assessed using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). The cross-validation results demonstrated that Random Forest achieved the most robust predictive performance for both cost overruns (R² = 0.912, RMSE = 1.685) and schedule delays (R² = 0.894, RMSE = 3.236). Furthermore, SHAP analysis identified human-related factors as the primary drivers of cost overruns and managerial factors as the dominant contributors to schedule delays. The proposed framework provides an interpretable decision-support tool for proactive safety management, early risk identification, and improved planning of irrigation projects.
Cost overrun is one of the main problems in project management that has a significant impact on the success of construction projects. High project complexity, uncertainty of field conditions, design changes, and external factors such as inflation and supply chain disruptions make predicting cost overruns a challenge that requires a more sophisticated analytical approach than conventional methods. The development of Machine Learning (ML) technology has opened new opportunities to improve the accuracy of project cost predictions through the use of historical data and complex pattern analysis. This study aims to review the development of research related to the application of Machine Learning in predicting project cost overruns, identify the most widely used algorithms, influential predictor variables, and evaluate the advantages and limitations of each approach. The research method is carried out through a systematic literature review of scientific publications indexed by Scopus in the period 2015–2025. The study focuses on various ML algorithms, including Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Gradient Boosting, and Extreme Gradient Boosting (XGBoost), which are used in project cost prediction. The study results show that Random Forest and XGBoost to produce higher levels of accuracy than traditional statistical methods or single ML models. This study concludes that Machine Learning has great potential in supporting early warning systems and data-driven decision-making to control the risk of project cost overruns.
J. Anggawan, Agustinus Purna Irawan, M. Waty· JURNAL TEKNIK SIPIL CENDEKIA...· 0 citations
Schedule delay remains a persistent challenge in large-scale infrastructure programmes, yet many predictive studies prioritise model accuracy while offering limited support for managerial interpretation and decision-making. This study aims to develop and evaluate a case-grounded, explainable hybrid machine-learning framework for schedule-delay risk prediction and decision-support translation.
The study integrates random forest and support vector machine within a weighted ensemble, with genetic algorithm optimisation applied to the random forest configuration. Shapley additive explanations is used to interpret model-attributed predictive contributions and identify non-linear risk-state patterns. The framework uses 470 case-grounded, programme-level risk-state observations derived from screened expert assessments and cross-checked against project documentation, practitioner interviews and schedule records.
The random forest-genetic algorithm + support vector machine model produced strong predictive results within the studied data set, achieving 92.2% accuracy with balanced precision, recall and F1-score values. The explainability analysis indicates that predicted schedule-delay risk is associated with non-linear, model-attributed risk-state patterns rather than isolated variables. Material and equipment supply, contractor performance and design changes emerged as the most salient predictive signals. These signals represent model-attributed contributions to predicted delay probability, not causal effects.
The study is based on a single large-scale infrastructure programme; therefore, broader transferability requires validation across multiple programmes, delivery systems and national contexts. The data are partly expert-informed, although screened and cross-checked against documentary and schedule evidence. SHAP-based explanations are model-attributed and should not be interpreted as causal effects. Future research should examine temporal modelling and benchmarking against XGBoost, LightGBM, CatBoost and deep-learning approaches.
The framework supports structured managerial prioritisation by translating model outputs into operational risk bands, trigger conditions, responsible actors, intervention timing and measurable outcomes.
By translating delay-risk predictions into explainable and trigger-based decision-support actions, the framework can support more transparent and accountable infrastructure project governance. Earlier identification of schedule-risk conditions may help reduce avoidable delays, resource waste, contractual disputes and disruption to public-service delivery. In large infrastructure projects, improved schedule-risk management can contribute to more reliable delivery of assets that affect communities, economic activity and public-sector investment efficiency. However, the framework should support, not replace, professional judgement and stakeholder accountability.
The study contributes by integrating hybrid prediction, explainable artificial intelligence and operational decision-support translation within a case-grounded infrastructure programme context. Rather than treating predictive accuracy or SHAP rankings as final outputs, the framework links predicted delay probability, dominant model-attributed risk signals and operational domains to trigger-based decision support.
Omid Tasa, M. Golabchi, M. Ravanshadnia· Journal of Engineering, Desi...· 0 citations
This study presents initial results on the integration of machine learning into a sustainability decision support system (MLDSS) designed to enhance organizational sustainability performance management. The proposed framework combines rule-based compliance assessment, ESG scoring, progress measurement, and strategic recommendations to support companies in aligning sustainability goals with operational priorities. A random forest regressor was employed to predict ESG scores from sustainability data extracted from corporate reports, focusing on greenhouse gas emission indicators. Experiments conducted with datasets from 5, 10, and 15 companies demonstrated that model performance improved with larger sample sizes, as reflected in rising R² (0.64 - 0.86) and explained variance scores, alongside reduced error margins. Meanwhile, the RMSE between the actual and model output for the GHGs Emission Score is reduced from 18.58 to 14.6 .The smaller datasets revealed limitations due to imbalance and variability, results from the 15-companies sample size indicated promising predictive capability and underscored the importance of continuous data enrichment. Beyond technical performance, the system’s visualization and tracking tools facilitate transparent stakeholder communication, reinforcing consumer trust and investor confidence. These findings highlight the potential of machine learning–assisted DSS to automate compliance, optimize sustainability strategies, and strengthen ESG reporting, offering a scalable pathway for organizations to advance sustainable practices.
Climate change poses a risk to sustainable agriculture through interference with irrigation timing, crop production, and resource management. Combining Artificial Intelligence with MPC and IoT can offer a basis for building climate-resilient decision support systems for precision agriculture. This study presents a Systematic Literature Review (SLR) of research published between 2019 and 2025 that investigates the synergistic use of Long Short-Term Memory LSTM neural networks, MPC, and IoT architectures for adaptive agricultural management. Following PRISMA and Kitchenham guidelines, 147 studies were screened, and 28 were analyzed using a ten-criterion Quality Assessment Checklist (QAC) with Weighted Quality Score (WQS) evaluation. Findings indicate that LSTM-based climate prediction models achieve average R² > 0.9, while MPC enhances irrigation efficiency by up to 30% under uncertain conditions. IoT infrastructures enable real-time communication and feedback across distributed farm networks, reducing latency and operational costs. The synthesis highlights an emerging shift toward closed-loop predictive–adaptive architectures. Despite significant progress, real-time field deployment remains limited due to computational and interoperability challenges. This SLR proposes a conceptual framework uniting LSTM forecasting, MPC optimization, and IoT-based monitoring for sustainable climate-resilient precision agriculture.
Ramadan Ahmad, Abdulkarim Mohammed· AlQalam journal of medical a...· 0 citations
Agricultural decision-support systems powered by machine learning (ML) have emerged as critical tools for enhancing crop productivity and food security, particularly for smallholder farmers in developing regions. However, many existing models lack methodological rigor, statistical validation, and interpretability—key requirements for real-world deployment. This study presents a comprehensive, reproducible ML framework for crop recommendation that integrates comparative feature selection, multi-model benchmarking, statistical significance testing with Nadeau–Bengio correction, and SHAP-based explainability. Using a dataset of 10,000 agricultural records from South Asia comprising 36 original features reduced to 10 selected predictors, we evaluated seven base classifiers alongside two ensemble strategies. Random Forest achieved the highest fivefold cross-validated accuracy (79.8%, 95% CI [79.0–80.6%]) and Macro F1 (0.474, 95% CI [0.455–0.492]), with consistent performance under tenfold cross-validation (accuracy: 79.7%, 95% CI [79.1–80.4%]; Macro F1: 0.471, 95% CI [0.460–0.482]). The Friedman test confirmed significant differences among models (χ2 = 25.25, df = 7, p < 0.001). Nadeau–Bengio corrected post-hoc testing revealed that Random Forest significantly outperformed only SVM and Logistic Regression (p < 0.05), while differences with gradient-boosting models were statistically non-significant in fivefold CV, though tenfold analysis showed marginal significance against Gradient Boosting (p = 0.0243). SHAP analysis identified rainfall (mean |SHAP|= 0.0772), temperature (0.0613), soil pH (0.0300), and nitrogen content (0.0244) as the top predictive features. Regional analysis revealed crop-specific optimal conditions: Rice (pH 5.48–7.43, rainfall 2209 mm), Wheat (pH 6.39–7.79, rainfall 1470 mm), Maize (pH 5.13–6.89, rainfall 1048 mm), and Cotton (pH 5.07–6.63, rainfall 594 mm). All code, data, and artifacts are publicly available. While pronounced class imbalance inherently constrains minority-class prediction, this framework establishes methodological standards for reproducible, interpretable agricultural AI by prioritizing statistical validation with proper dependence correction, imbalance-aware evaluation, and agronomic explainability.
Nigeria's natural gas sector loses an estimated USD 2.5 billion annually to suboptimal utilisation and persistent flaring, yet conventional forecasting tools remain inadequate for the sector's nonlinear volatility. This study develops and validates an integrated machine learning framework for forecasting monthly natural gas utilisation and revenue using 48 months of official NNPC production data (January 2021 – December 2024). Five models are benchmarked on a 12-month hold-out test set: SARIMA as the statistical baseline, Random Forest, XGBoost, a two-layer LSTM network, and a Hybrid Ensemble combining the three ML learners via inverse-RMSE weighting. All ML models exceeded the target of a 15% RMSE reduction over SARIMA, with the Hybrid Ensemble achieving the greatest improvement at approximately 48.8% and an R2 of 0.93. SHAP feature importance analysis identified lagged utilisation and rolling-mean features as the dominant predictors, followed by upstream production volumes and cyclical seasonal patterns. Retrained on the full dataset, the Hybrid Ensemble generates a 60-month forecast (2025–2029) across three scenarios: a Base Case projecting stable annual utilisation of approximately 2,320 BSCF/year and cumulative revenues of approximately USD 13.0 billion, and an Optimistic scenario projecting 2,795 BSCF/year and USD 31.4 billion under favourable LNG pricing and capacity expansion. Flaring reduction modelling indicates that achieving a 22% reduction in the current 7.7% flaring rate could recover approximately USD 0.93 billion in otherwise wasted gas value over five years, with co-benefits aligned to Nigeria's 2030 zero-flaring pledge and Nationally Determined Contributions. These findings offer a replicable, interpretable decision-support framework for operators, regulators, and investors seeking to align gas production forecasts with Nigeria's energy transition objectives.
S. Dauda, A. Bakare, P. O. Enarevba et al.· SPE Nigeria Annual Internati...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
