Mohsen Mohammadagha

This study addresses the pressing challenge of supervised face recognition under unconstrained conditions by systematically integrating classical machine learning, deep learning, and transfer learning approaches. While existing literature demonstrates significant progress, particularly with hybrid and transfer learning models, a gap remains in unified, detailed benchmarking across diverse techniques. The primary objective is to holistically compare Support Vector Machines (SVM) with PCA, Artificial Neural Networks (ANN), XGBoost, a custom Convolutional Neural Network (CNN), MobileNetV2-based transfer learning, and stacked hybrid meta-models using the Labeled Faces in the Wild (LFW) dataset. The methodology encompasses data preprocessing, parallel feature extraction, dimensionality reduction, ensemble learning, and interpretability analysis. Experimental results show that stacking hybrid models achieves the highest test accuracy (87.9%) and macro ROC-AUC (0.983), with MobileNetV2 transfer learning also excelling in sample efficiency and performance. Future research should expand interpretability diagnostics and benchmark these pipelines on more diverse or occluded datasets for greater real-world applicability.

Polyvinyl chloride (PVC) accounts for 23%–27% of installed sewer pipe, yet utilities increasingly need data-driven tools to complement inspection-based assessments by reducing subjectivity and enabling consistent prioritization. To address these limitations, this study develops an automated, objective condition classification and risk-based maintenance planning by developing and comparing three machine-learning classifiers, Random Forest (RF), Gradient Boosting (GB), and Artificial Neural Networks (ANN), trained on 1,400 PVC pipe segments from Tampa, Florida, using PACP condition grades (1–5) as the target variable. A multi-model framework including preprocessing, feature engineering, and hyperparameter tuning was implemented. The RF achieved the best overall performance (accuracy 96.1%, F1-score 94.4%), while ANN provided stronger class discrimination for some grades, area under the receiver operating characteristic curve, (ROC-AUC) up to 0.81, indicating sensitivity to class imbalance. Feature importance was consistent across models, with age and length contributing 42% and 30%, respectively. Future research should incorporate imbalance-aware learning, multi-utility validation, and additional hydraulic/groundwater covariates and develop longitudinal deterioration forecasting for renewal decision support.

Underground 15‑kV XLPE cable networks are critical to distribution reliability yet degrade through aging, partial discharge (PD), and corrosion, motivating data‑driven condition assessment beyond periodic manual inspection. This study addresses the need for scalable and explainable models on real utility inspection records by developing an interpretable multi‑model stacking ensemble to classify categorical visual condition states. Using 2,500 asset‑level cable records from Western Canada, four standardized predictors (Age, Partial_Discharge, Neutral_Corrosion, Loading) were analyzed and modeled with Logistic Regression, Gradient Boosting, Random Forest, and a stacking ensemble with a Random Forest meta‑learner. The stacking ensemble achieved the best overall performance, with accuracy 95.80%, precision 95.79%, recall 95.80%, and F1‑score 95.72%, while feature-importance and SHAP-based explanations consistently identified Age and PD as dominant degradation drivers. Future work should validate transferability across multiple utilities, incorporate temporal PD/loading trajectories, quantify uncertainty and drift for deployment, and integrate model outputs into EAM/GIS workflows for operational decision support.

Extractive question answering on lengthy documents remains computationally expensive due to quadratic attention complexity and context truncation requirements in modern language models. This work proposes a hybrid context filtering framework that combines classical similarity metrics, including cosine similarity and Word Mover’s Distance, with the Bitap algorithm, and utilizes selective LLM-based validation to reduce inference cost while maintaining competitive accuracy. The method filters irrelevant sentences before passage encoding, thereby reducing computational overhead without requiring learned retrieval components. Evaluation on SQuAD 2.0 across four open-source models (Llama 2 8B, T5-3B, Flan-T5-XL, mT5-Base) using 5-shot learning and fine-tuning demonstrates a 2.3 inference speedup and 58% latency reduction with a modest accuracy trade-off of 5.7% relative F1 degradation compared to full-context baselines. Component ablation confirms the synergistic contribution of each similarity metric, while robustness evaluation across various context lengths and out-of-distribution settings validates the method’s generalization capabilities. These results indicate that intelligent, parameter-free context filtering can achieve meaningful computational efficiency without necessitating complex learned retrievers.

This study evaluates the predictive power of linear and logistic regression models on the Iris dataset, emphasizing feature importance and model performance, and comparing with construction assessment. Machine learning's role in predictive analysis is explored, with the Iris dataset serving as a benchmark for classification and regression tasks. The research addresses the need for robust methodologies to enhance model accuracy and interpretability. The objectives include comparing linear, logistic regression, and ANN approaches to highlight their strengths and limitations. Methodologically, data preprocessing, feature scaling, and Python-based implementations were employed to ensure reliable outcomes. Results indicate that ANN outperforms MLR (93.33%) in this research due to its adaptability to nonlinear relationships, achieving higher accuracy (97%). This aligns with findings in construction assessment studies, where ANNs' advanced methodology also demonstrated superior performance over MLR. Future research should integrate advanced machine learning models such as Neural Architecture Search (NAS), multimodal approaches, and ensemble techniques—including bagging (e.g., Random Forest), boosting (e.g., AdaBoost, XGBoost), and stacking. Related methods, such as voting classifiers, blending, and mixture of experts, can further enhance feature selection, interpretability, and predictive performance across domains.

This study addresses the pressing challenge of supervised face recognition under unconstrained conditions by systematically integrating classical machine learning, deep learning, and transfer learning approaches. While existing literature demonstrates significant progress, particularly with hybrid and transfer learning models, a gap remains in unified, detailed benchmarking across diverse techniques. The primary objective is to holistically compare Support Vector Machines (SVM) with PCA, Artificial Neural Networks (ANN), XGBoost, a custom Convolutional Neural Network (CNN), MobileNetV2-based transfer learning, and stacked hybrid meta-models using the Labeled Faces in the Wild (LFW) dataset. The methodology encompasses data preprocessing, parallel feature extraction, dimensionality reduction, ensemble learning, and interpretability analysis. Experimental results show that stacking hybrid models achieves the highest test accuracy (87.9%) and macro ROC-AUC (0.983), with MobileNetV2 transfer learning also excelling in sample efficiency and performance. Future research should expand interpretability diagnostics and benchmark these pipelines on more diverse or occluded datasets for greater real-world applicability.

This paper introduces Lean 5.0, a human-centric evolution of Lean-Digital integration that connects predictive analytics, AI collaboration, and continuous learning within Industry 5.0 and Construction 5.0 contexts. A systematic literature review (2019-2024) and a 12-week empirical validation study demonstrate measurable performance gains, including a 13% increase in Plan Percent Complete (PPC), 22% reduction in rework, and 42% improvement in forecast accuracy. The study adopts a mixed-method Design Science Research (DSR) approach aligned with PRISMA 2020 guidelines. The paper also examines integration with digital twin and blockchain technologies to improve traceability, auditability, and lifecycle transparency. Despite limitations related to sample size, single-case design, and study duration, the findings show that Lean 5.0 provides a transformative paradigm connecting human cognition with predictive control in construction management.

This study addresses the persistent challenge of overfitting in deep learning-based image classification, emphasizing the need for robust regularization strategies as models become increasingly complex. To meet the research need for systematic, quantitative comparisons of regularization techniques across different architectures, the objective was to evaluate how methods such as dropout and data augmentation impact generalization in both baseline CNNs and ResNet-18 models. The methodology involved controlled experiments using the Imagenette dataset at varying resolutions, with consistent application of regularization, early stopping, and transfer learning protocols. Results show that ResNet-18 achieved superior validation accuracy (82.37%) compared to the baseline CNN (68.74%), and that regularization reduced overfitting and improved generalization across all scenarios. Transfer learning further enhanced performance, with fine-tuned models converging faster and attaining higher accuracy than those trained from scratch. Future research should explore the interplay between emerging regularization methods and novel architectures, as well as their effectiveness in transfer learning and resource-constrained environments, to further advance the reliability and efficiency of deep learning systems for image classification.